Reject OTP verify responses lacking a usable expiry

Accept epoch-seconds relay expiry so valid verification is not misread
Session-gate historical run reads; fail closed on missing expiry
2026-08-16 09:26:39 +02:00 · 2026-07-21 00:06:40 +00:00 · 2026-07-21 00:00:43 +00:00 · 2026-07-20 23:55:04 +00:00 · 2026-07-20 23:45:11 +00:00 · 2026-07-20 19:06:40 -04:00
146 changed files with 20071 additions and 328 deletions
@@ -1,17 +1,25 @@
+# Node / local-viewer SPA source (the built bundle in
+# strix/viewer/viewer_dist/ is committed and shipped; do not ignore it)
+node_modules/
+strix/viewer_src/node_modules/
+strix/viewer_src/.vite/
+
 # Python
 __pycache__/
 *.py[cod]
 *$py.class
 *.so
 .Python
-build/
+# Anchored to the repo root: these are Python build-artifact dir names, but
+# unanchored they also match nested source dirs (e.g. the viewer's src/lib).
+/build/
 develop-eggs/
 dist/
 downloads/
 eggs/
 .eggs/
-lib/
-lib64/
+/lib/
+/lib64/
 parts/
 sdist/
 var/
@@ -46,7 +54,7 @@ pip-delete-this-directory.txt
 .env.production.local

 # MongoDB
-data/
+/data/
 mongod.log
 *.mongodb
 *.mongorc.js
@@ -99,6 +99,20 @@ We welcome feature ideas! Please:
 - Consider implementation approach
 - Be open to discussion

+## 🖥️ Local viewer SPA
+
+`strix view` serves a prebuilt web UI whose source lives in `strix/viewer_src/`
+(a Vite + React project) and whose built output is committed to
+`strix/viewer/viewer_dist/` and shipped in the package. End users never run a
+JS build. If you change anything under `strix/viewer_src/`, rebuild and commit
+the output:
+
+```bash
+make viewer   # or: cd strix/viewer_src && npm ci && npm run build
+```
+
+Commit both the source change and the regenerated `strix/viewer/viewer_dist/`.
+
 ## 🤝 Community

 - **Discord**: [Join our community](https://discord.gg/strix-ai)
@@ -1,4 +1,4 @@
-.PHONY: help install dev-install format lint type-check security check-all clean pre-commit setup-dev dev
+.PHONY: help install dev-install format lint type-check security check-all clean pre-commit setup-dev dev viewer

 help:
 	@echo "Available commands:"
@@ -15,6 +15,7 @@ help:
 	@echo ""
 	@echo "Development:"
 	@echo "  pre-commit    - Run pre-commit hooks on all files"
+	@echo "  viewer        - Rebuild the local-viewer SPA (commit the output)"
 	@echo "  clean         - Clean up cache files and artifacts"

 install:
@@ -66,5 +67,10 @@ clean:
 	find . -name "*.pyc" -delete 2>/dev/null || true
 	@echo "✅ Cleanup complete!"

+viewer:
+	@echo "🖥️  Building the local-viewer SPA..."
+	cd strix/viewer_src && npm ci && npm run build
+	@echo "✅ Viewer built to strix/viewer/viewer_dist/ (commit the changes)."
+
 dev: format lint type-check
 	@echo "✅ Development cycle complete!"
@@ -27,8 +27,8 @@
 <a href="https://x.com/strix_ai"><img src="https://github.com/usestrix/.github/raw/main/imgs/X.png" height="40" alt="Follow on X"></a>


-<a href="https://trendshift.io/repositories/15362" target="_blank"><img src="https://trendshift.io/api/badge/repositories/15362" alt="usestrix/strix | Trendshift" width="250" height="55"/></a>
 <a href="https://trendshift.io/repositories/15362?utm_source=trendshift-badge&amp;utm_medium=badge&amp;utm_campaign=badge-trendshift-15362" target="_blank" rel="noopener noreferrer"><img src="https://trendshift.io/api/badge/trendshift/repositories/15362/weekly" alt="usestrix%2Fstrix | Trendshift" width="250" height="55"/></a>
+<a href="https://trendshift.io/repositories/15362" target="_blank"><img src="https://trendshift.io/api/badge/repositories/15362" alt="usestrix/strix | Trendshift" width="250" height="55"/></a>

 </div>

@@ -24,6 +24,7 @@ RUN apt-get update && \
    python3 python3-pip python3-dev python3-venv python3-setuptools \
    golang-go \
    net-tools dnsutils whois \
+    file xxd \
    jq parallel ripgrep grep \
    less man-db procps htop \
    iproute2 iputils-ping netcat-traditional \
@@ -192,6 +193,8 @@ RUN mkdir -p /workspace && chown -R pentester:pentester /workspace /app
 USER pentester
 RUN python3 -m venv /app/.venv && \
    /app/.venv/bin/pip install --no-cache-dir caido-sdk-client && \
+    /app/.venv/bin/pip install --no-cache-dir \
+        requests httpx beautifulsoup4 lxml pyjwt cryptography && \
    /app/.venv/bin/pip install --no-cache-dir -r /home/pentester/tools/jwt_tool/requirements.txt && \
    printf '%s\n' \
        '#!/bin/bash' \
@@ -91,10 +91,13 @@ http_proxy=http://127.0.0.1:${CAIDO_PORT}
 https_proxy=http://127.0.0.1:${CAIDO_PORT}
 EOF

-echo "source /etc/profile.d/proxy.sh" >> ~/.bashrc
-echo "source /etc/profile.d/proxy.sh" >> ~/.zshrc
+# Use POSIX `.` (not the bashism `source`) so these lines are safe when the rc
+# files are read by a POSIX shell (e.g. `sh -lc`), which otherwise fails with
+# "source: not found". `.` is understood by bash, zsh, and dash alike.
+echo ". /etc/profile.d/proxy.sh" >> ~/.bashrc
+echo ". /etc/profile.d/proxy.sh" >> ~/.zshrc

-source /etc/profile.d/proxy.sh
+. /etc/profile.d/proxy.sh

 echo "✅ System-wide proxy configuration complete"

@@ -81,6 +81,14 @@ Protocol-specific testing techniques.
 | --------- | ------------------------------------------------ |
 | `graphql` | GraphQL introspection, batching, resolver issues |

+### Reconnaissance
+
+Passive discovery and attack-surface mapping techniques.
+
+| Skill             | Coverage                                                        |
+| ----------------- | --------------------------------------------------------------- |
+| `asset_discovery` | CT, TLS SAN pivoting, passive DNS, and ASN/IP asset enumeration |
+
 ### Tooling

 Sandbox CLI playbooks for core recon and scanning tools.
@@ -1,6 +1,6 @@
 [project]
 name = "strix-agent"
-version = "1.0.4"
+version = "1.1.0"
 description = "Open-source AI Hackers for your apps"
 readme = "README.md"
 license = "Apache-2.0"
@@ -34,6 +34,8 @@ classifiers = [
 ]
 dependencies = [
  "openai-agents[litellm]==0.14.6",
+  "openai>=2.26.0,<2.45",
+  "litellm",
  "pydantic>=2.11.3",
  "pydantic-settings>=2.13.0",
  "rich",
@@ -42,6 +44,9 @@ dependencies = [
  "requests>=2.32.0",
  "cvss>=3.2",
  "caido-sdk-client>=0.2.0",
+  "reportlab>=4.0",
+  "pypdf>=5.0",
+  "cryptography>=42",
 ]

 [project.optional-dependencies]
@@ -72,6 +77,10 @@ build-backend = "hatchling.build"

 [tool.hatch.build.targets.wheel]
 packages = ["strix"]
+# The prebuilt viewer bundle under strix/viewer/viewer_dist/ ships automatically
+# (hatchling includes non-.py files under the package). The Vite SOURCE lives
+# under the package dir too but must never ship in the wheel.
+exclude = ["strix/viewer_src", "strix/viewer_src/**"]

 # ============================================================================
 # Type Checking Configuration
@@ -109,6 +118,8 @@ module = [
    "docker.*",
    "caido_sdk_client.*",
    "pydantic_settings.*",
+    "reportlab.*",
+    "pypdf.*",
 ]
 ignore_missing_imports = true
 disable_error_code = ["import-untyped"]
@@ -199,6 +210,15 @@ ignore = [
 ]

 [tool.ruff.lint.per-file-ignores]
+# Test doubles use fixture tokens/passwords and match a callee signature whose
+# args they intentionally ignore.
+"tests/test_viewer_auth.py" = ["S105", "S106", "ARG001"]
+"tests/test_report_pdf.py" = ["S105", "S106"]
+# Stdlib HTTP handler overrides (do_GET/do_POST) and lazy imports that avoid a
+# circular dependency with strix.telemetry / strix.viewer.report_pdf.
+"strix/viewer/server.py" = ["N802", "PLC0415"]
+# Lazy telemetry import to avoid importing PostHog before the viewer starts.
+"strix/viewer/cli.py" = ["PLC0415"]
 # Lazy imports inside functions to avoid circular dependency with
 # strix.telemetry / strix.report.dedupe / cvss.
 "strix/tools/notes/tools.py" = ["PLC0415", "TC002"]
@@ -25,6 +25,13 @@ for tcss_file in strix_root.rglob('*.tcss'):
    rel_path = tcss_file.relative_to(project_root)
    datas.append((str(tcss_file), str(rel_path.parent)))

+# Prebuilt local-viewer SPA (served by `strix view`).
+viewer_dist = strix_root / 'viewer' / 'viewer_dist'
+for asset in viewer_dist.rglob('*'):
+    if asset.is_file():
+        rel_path = asset.relative_to(project_root)
+        datas.append((str(asset), str(rel_path.parent)))
+
 datas += collect_data_files('textual')

 datas += collect_data_files('tiktoken')
@@ -151,6 +158,21 @@ hiddenimports = [
    'strix.report.dedupe',
    'strix.report.state',
    'strix.report.writer',
+    'strix.viewer',
+    'strix.viewer.auth',
+    'strix.viewer.cli',
+    'strix.viewer.report_pdf',
+    'strix.viewer.server',
+    'strix.viewer.transcript',
+
+    # PDF report generation + encryption
+    'reportlab',
+    'reportlab.pdfgen',
+    'reportlab.pdfbase',
+    'reportlab.lib',
+    'reportlab.platypus',
+    'pypdf',
+    'cryptography',
    'strix.runtime',
    'strix.runtime.backends',
    'strix.runtime.caido_bootstrap',
@@ -178,6 +200,16 @@ hiddenimports += collect_submodules('textual')
 hiddenimports += collect_submodules('rich')
 hiddenimports += collect_submodules('pydantic')
 hiddenimports += collect_submodules('pygments')
+# reportlab loads renderers/fonts dynamically, so pull its whole tree in.
+hiddenimports += collect_submodules('reportlab')
+
+# reportlab ships bundled fonts (.pfb/.afm) it needs at runtime.
+datas += collect_data_files('reportlab')
+
+# reportlab imports PIL (pillow) lazily for image handling, so it must be
+# bundled explicitly and kept out of the excludes list below.
+hiddenimports += collect_submodules('PIL')
+datas += collect_data_files('PIL')

 excludes = [
    # Sandbox-only packages
@@ -225,7 +257,6 @@ excludes = [
    'numpy',
    'pandas',
    'scipy',
-    'PIL',
    'cv2',
 ]

@@ -41,7 +41,7 @@ from strix.tools.proxy.tools import (
    view_request,
    view_sitemap_entry,
 )
-from strix.tools.reporting.tool import create_vulnerability_report
+from strix.tools.reporting.tool import create_dependency_report, create_vulnerability_report
 from strix.tools.thinking.tool import think
 from strix.tools.todo.tools import (
    create_todo,
@@ -55,7 +55,7 @@ from strix.tools.web_search.tool import web_search


 if TYPE_CHECKING:
-    from collections.abc import Awaitable, Callable
+    from collections.abc import Awaitable, Callable, Sequence

    from agents import RunContextWrapper
    from agents.tool import FunctionToolResult
@@ -209,6 +209,13 @@ def _wrap_exec_command(tool: FunctionTool) -> FunctionTool:
    invoke_tool = tool.on_invoke_tool

    async def invoke(ctx: Any, raw_input: str) -> Any:
+        try:
+            parsed = json.loads(raw_input)
+        except (json.JSONDecodeError, TypeError):
+            parsed = None
+        if isinstance(parsed, dict) and "shell" not in parsed:
+            parsed["shell"] = "bash"
+            raw_input = json.dumps(parsed)
        try:
            return await invoke_tool(ctx, raw_input)
        except ValidationError as exc:
@@ -335,6 +342,7 @@ _BASE_TOOLS: tuple[Tool, ...] = (
    delete_note,
    web_search,
    create_vulnerability_report,
+    create_dependency_report,
    list_requests,
    view_request,
    repeat_request,
@@ -349,6 +357,48 @@ _BASE_TOOLS: tuple[Tool, ...] = (
 )


+# Extra tools registered for scan agents. Mirrors
+# ``strix.runtime.backends.register_backend``: register before the first
+# ``build_strix_agent`` call and every agent (root + children) gets them.
+_EXTRA_TOOLS: list[Tool] = []
+
+
+def _ensure_unique_tool_names(tools: Sequence[Tool]) -> None:
+    seen: set[str] = set()
+    duplicates: set[str] = set()
+    for tool in tools:
+        if tool.name in seen:
+            duplicates.add(tool.name)
+        seen.add(tool.name)
+    if duplicates:
+        msg = f"Agent tools must have unique names: {sorted(duplicates)}"
+        raise ValueError(msg)
+
+
+def register_agent_tools(*tools: Tool) -> None:
+    """Register tools for every scan agent built afterwards.
+
+    Tools are added to both root and child agents, after the base set and
+    before the lifecycle tool (``finish_scan`` / ``agent_finish``). Duplicate
+    tool objects are ignored so repeated imports don't double-register.
+    """
+    new_tools: list[Tool] = []
+    for tool in tools:
+        if tool not in _EXTRA_TOOLS and tool not in new_tools:
+            new_tools.append(tool)
+
+    _ensure_unique_tool_names([*_BASE_TOOLS, *_EXTRA_TOOLS, *new_tools, finish_scan, agent_finish])
+
+    for tool in new_tools:
+        _EXTRA_TOOLS.append(tool)
+        logger.info("Registered extra agent tool: %s", getattr(tool, "name", tool))
+
+
+def registered_agent_tools() -> tuple[Tool, ...]:
+    """Return the currently registered scan-agent tools."""
+    return tuple(_EXTRA_TOOLS)
+
+
 def build_strix_agent(
    *,
    name: str = "strix",
@@ -359,26 +409,37 @@ def build_strix_agent(
    interactive: bool = False,
    chat_completions_tools: bool = False,
    system_prompt_context: dict[str, Any] | None = None,
+    extra_tools: Sequence[Tool] | None = None,
+    instructions_override: str | None = None,
 ) -> SandboxAgent[Any]:
    """Build a SandboxAgent for either root or child use.

    Args:
        chat_completions_tools: Wrap SDK custom tools as function tools
            when the selected backend cannot accept Responses custom tools.
+        extra_tools: Additional tools for this scan agent only, on top of any
+            registered via ``register_agent_tools``.
+        instructions_override: Use this verbatim as the system prompt instead
+            of rendering the built-in scan prompt.
    """
-    instructions = render_system_prompt(
-        skills=skills,
-        scan_mode=scan_mode,
-        is_whitebox=is_whitebox,
-        is_root=is_root,
-        interactive=interactive,
-        system_prompt_context=system_prompt_context,
-    )
-
-    if is_root:
-        tools: list[Tool] = [*_BASE_TOOLS, finish_scan]
+    if instructions_override is not None:
+        instructions = instructions_override
    else:
-        tools = [*_BASE_TOOLS, agent_finish]
+        instructions = render_system_prompt(
+            skills=skills,
+            scan_mode=scan_mode,
+            is_whitebox=is_whitebox,
+            is_root=is_root,
+            interactive=interactive,
+            system_prompt_context=system_prompt_context,
+        )
+
+    agent_tools = [*_EXTRA_TOOLS, *(extra_tools or [])]
+    if is_root:
+        tools: list[Tool] = [*_BASE_TOOLS, *agent_tools, finish_scan]
+    else:
+        tools = [*_BASE_TOOLS, *agent_tools, agent_finish]
+    _ensure_unique_tool_names(tools)

    logger.info(
        "Built %s agent '%s' (skills=%d, tools=%d, scan_mode=%s, whitebox=%s)",
@@ -7,7 +7,7 @@ from typing import Any

 from jinja2 import Environment, FileSystemLoader, select_autoescape

-from strix.skills import get_available_skills, load_skills
+from strix.skills import get_available_skills, load_skills, skill_search_dirs
 from strix.utils.resource_paths import get_strix_resource_path


@@ -69,9 +69,9 @@ def render_system_prompt(
    """Render the system prompt. Returns empty string on template failure."""
    try:
        prompt_dir = get_strix_resource_path("agents", _PROMPT_DIRNAME)
-        skills_dir = get_strix_resource_path("skills")
+        loader_dirs = [prompt_dir, *skill_search_dirs()]
        env = Environment(
-            loader=FileSystemLoader([prompt_dir, skills_dir]),
+            loader=FileSystemLoader(loader_dirs),
            autoescape=select_autoescape(
                enabled_extensions=(),
                default_for_string=False,
@@ -168,9 +168,24 @@ EFFICIENCY TACTICS:
 - Download additional tools as needed for specific tasks
 - Run multiple scans in parallel when possible
 - Load the most relevant skill before starting a specialized testing workflow if doing so will improve accuracy, speed, or tool usage
- Use `exec_command` for Python code: write reusable scripts under
-  `/workspace/scratch/` and run them with `python3`. For one-off snippets,
-  `python3 -c` or a here-document is acceptable.
+- Use `exec_command` for Python code: write reusable scripts to a file and
+  run them with `python3 script.py`. For one-off snippets, `python3 -c` or a
+  here-document is acceptable, but avoid deeply nested quotes/parentheses — if
+  a snippet needs complex quoting or is more than a few lines, write it to a
+  file first to prevent syntax errors.
+- Before importing a third-party Python library, make sure it is installed. The
+  sandbox's `python3` runs inside a preconfigured virtualenv that ships
+  `requests`, `httpx`, `beautifulsoup4` (bs4), `lxml`, `pyjwt`, and
+  `cryptography`; for anything else prefer the stdlib or run `pip install <pkg>`
+  (it installs into that active venv) before importing, rather than letting the
+  script fail with `ModuleNotFoundError`.
+- `exec_command` runs each command in a fresh non-interactive shell (plain
+  pipes, no TTY). To drive an interactive or long-running process with
+  `write_stdin` — REPLs, `ssh`/`nc`/`ftp`, `msfconsole`, or to send Ctrl-C —
+  you MUST start it with `exec_command(cmd="...", tty=true)` and then
+  `write_stdin(session_id=<id>, chars="...")`. Calling `write_stdin` on a
+  default (non-TTY) command or on a process that has already exited fails with
+  "stdin is not available".
 - For Caido proxy automation inside Python, explicitly import from
  `caido_api`:
  `from caido_api import list_requests, view_request, repeat_request, list_sitemap, view_sitemap_entry, scope_rules`
@@ -186,11 +201,11 @@ EFFICIENCY TACTICS:
 VALIDATION REQUIREMENTS:
 - Full validation required - no assumptions
 - Demonstrate concrete impact with evidence
- Consider business context for severity assessment
+- Consider business context for severity assessment — check whether the target is a demo/sandbox environment or content meant to be public, and factor that in
 - Independent verification through subagent
 - Document complete attack chain
 - Keep going until you find something that matters
- A vulnerability is ONLY considered reported when a reporting agent uses create_vulnerability_report with full details. Mentions in agent_finish, finish_scan, or generic messages are NOT sufficient
+- A vulnerability is ONLY considered reported when a reporting agent uses create_vulnerability_report (or create_dependency_report for known-CVE dependency/supply-chain findings) with full details. Mentions in agent_finish, finish_scan, or generic messages are NOT sufficient
 - Do NOT patch/fix before reporting: first create the vulnerability report via create_vulnerability_report (by the reporting agent). Only after reporting is completed should fixing/patching proceed
 - DEDUPLICATION: The create_vulnerability_report tool uses LLM-based deduplication. If it rejects your report as a duplicate, DO NOT attempt to re-submit the same vulnerability. Accept the rejection and move on to testing other areas. The vulnerability has already been reported by another agent
 </execution_guidelines>
@@ -240,12 +255,18 @@ AGENT ISOLATION & SANDBOXING:
 - All agents share the same /workspace directory and proxy history
 - Agents can see each other's files and proxy traffic for better collaboration

+DISK & SCRATCH HYGIENE:
+- /workspace is a shared, finite disk used by all agents at once — be a considerate tenant
+- Prefer bounded recon: scope crawls and scans by depth, duration, and target rather than "collect everything"
+- Redirect large tool output to a file, and once you've extracted what you need (e.g. a URL/endpoint list), remove the raw output
+- If disk gets tight or a write fails for space, check what's large under /workspace and clean up files from your own task; leave another agent's files unless you've confirmed they're no longer in use
+
 MANDATORY INITIAL PHASES:

 BLACK-BOX TESTING - PHASE 1 (RECON & MAPPING):
 - COMPLETE full reconnaissance: subdomain enumeration, port scanning, service detection
 - MAP entire attack surface: all endpoints, parameters, APIs, forms, inputs
- CRAWL thoroughly: spider all pages (authenticated and unauthenticated), discover hidden paths, analyze JS files
+- CRAWL thoroughly: spider all pages (authenticated and unauthenticated), discover hidden paths, analyze JS files — keep each crawl bounded by depth/duration, and tidy up raw output once endpoints are extracted
 - ENUMERATE technologies: frameworks, libraries, versions, dependencies
 - Reconnaissance should normally happen before targeted vulnerability discovery unless the correct next move is already obvious or the user/system explicitly asks to prioritize a specific area first
 - ONLY AFTER comprehensive mapping → proceed to vulnerability testing
@@ -413,6 +434,7 @@ SPECIALIZED TOOLS:
 PROXY & INTERCEPTION:
 - Caido CLI - Modern web proxy (already running). Use the proxy tools
  directly, or import `caido_api` from sandbox Python scripts.
+- HTTPQL filters (for `list_requests`): quote string values, leave integers unquoted (`resp.code.eq:200`, not `"200"`); combine terms with `AND`/`OR` (there is no `NOT` — use the negated operator `ne`/`ncont`/`nregex`). Numeric fields (`resp.code`, `req.port`) use `eq`/`ne`/`gt`/`gte`/`lt`/`lte`; text fields (`req.host`, `req.path`, `req.method`, `req.raw`) use `cont`/`ncont`/`eq`/`regex`. Example: `resp.code.gte:200 AND resp.code.lt:300 AND req.host.cont:"api"`.
 - NOTE: If you are seeing proxy errors when sending requests, it usually means you are not sending requests to a correct url/host/port.
 - Ignore Caido proxy-generated 50x HTML error pages; these are proxy issues (might happen when requesting a wrong host or SSL/TLS issues, etc).

@@ -10,6 +10,7 @@ from agents.models.multi_provider import MultiProvider
 from agents.retry import (
    ModelRetryBackoffSettings,
    ModelRetrySettings,
+    RetryPolicyContext,
    retry_policies,
 )

@@ -20,6 +21,21 @@ if TYPE_CHECKING:
    from strix.config.settings import Settings


+def request_timeout_extra_args(timeout_s: float | None) -> dict[str, float] | None:
+    """Per-request model timeout; a plain float so ``ModelSettings.to_json_dict()`` stays serializable."""  # noqa: E501
+    if not timeout_s or timeout_s <= 0:
+        return None
+    return {"timeout": timeout_s}
+
+
+def _retry_statusless_provider_errors(context: RetryPolicyContext) -> bool:
+    """Retry statusless provider errors (e.g. mid-stream quota/billing), but not aborts."""
+    normalized = context.normalized
+    if normalized.is_abort:
+        return False
+    return normalized.status_code is None
+
+
 class StrixProvider(MultiProvider):
    """Route any non-OpenAI prefix through LiteLLM with the prefix preserved,
    so users type ``deepseek/deepseek-chat`` rather than
@@ -56,15 +72,53 @@ DEFAULT_MODEL_RETRY = ModelRetrySettings(
        retry_policies.provider_suggested(),
        retry_policies.network_error(),
        retry_policies.http_status((429, 500, 502, 503, 504)),
+        _retry_statusless_provider_errors,
    ),
 )

+RECOMMENDED_MODEL_NAMES = (
+    "openai/gpt-5.6",
+    "openai/gpt-5.6-sol",
+    "openai/gpt-5.6-terra",
+    "openai/gpt-5.5",
+    "openai/gpt-5.5-pro",
+    "openai/gpt-5.4",
+    "openai/gpt-5.3-codex",
+    "anthropic/claude-fable-5",
+    "anthropic/claude-opus-4-8",
+    "anthropic/claude-opus-4-7",
+    "anthropic/claude-sonnet-5",
+    "anthropic/claude-sonnet-4-6",
+    "vertex_ai/gemini-3.1-pro-preview",
+    "gemini/gemini-3.1-pro-preview",
+    "deepseek/deepseek-v4-pro",
+    "deepseek/deepseek-v4-flash",
+    "dashscope/qwen3.7-max-2026-06-08",
+    "moonshot/kimi-k2.7-code",
+    "moonshot/kimi-k2.6",
+)
+
+_RECOMMENDED_MODEL_NAME_SET = frozenset(name.lower() for name in RECOMMENDED_MODEL_NAMES)
+
+FRONTIER_MODEL_FAMILIES = (
+    (("azure", "azure_ai", "bedrock_mantle", "openai"), ("gpt-5",)),
+    (
+        ("anthropic", "azure_ai", "bedrock", "claude", "databricks", "snowflake", "vertex_ai"),
+        ("claude-fable-5", "claude-opus-4", "claude-sonnet-5", "claude-sonnet-4"),
+    ),
+    (("google", "gemini", "vertex_ai"), ("gemini-3",)),
+    (("deepseek",), ("deepseek-v4", "deepseek-r1", "deepseek-reasoner")),
+    (("alibaba", "dashscope", "qwen"), ("qwen3.7", "qwen3.5", "qwen3-max")),
+    (("moonshot", "moonshotai", "kimi"), ("kimi-k2.7", "kimi-k2.6", "kimi-k2.5")),
+)
+

 def configure_sdk_model_defaults(settings: Settings) -> None:
    """Apply Strix config to SDK-native defaults."""
    llm = settings.llm
    set_tracing_disabled(True)
    _configure_litellm_compatibility()
+    _configure_openrouter_attribution(llm.model)
    if llm.api_key:
        set_default_openai_key(llm.api_key, use_for_tracing=False)
        _configure_litellm_default("api_key", llm.api_key)
@@ -111,6 +165,29 @@ def _configure_litellm_compatibility() -> None:
    _register_litellm_cost_callback()


+_OPENROUTER_ATTRIBUTION_HEADERS = {
+    "HTTP-Referer": "https://strix.ai",
+    "X-Title": "Strix",
+    "X-OpenRouter-Categories": "cli-agent",
+}
+
+
+def _configure_openrouter_attribution(model_name: str | None) -> None:
+    import litellm
+
+    current: object = litellm.headers
+    existing: dict[str, str] = current if isinstance(current, dict) else {}
+    if not model_name or "openrouter/" not in model_name.strip().lower():
+        if any(key in existing for key in _OPENROUTER_ATTRIBUTION_HEADERS):
+            remaining = {
+                k: v for k, v in existing.items() if k not in _OPENROUTER_ATTRIBUTION_HEADERS
+            }
+            litellm.headers = remaining or None  # type: ignore[assignment]
+        return
+
+    litellm.headers = {**existing, **_OPENROUTER_ATTRIBUTION_HEADERS}  # type: ignore[assignment]
+
+
 def _register_litellm_cost_callback() -> None:
    import litellm

@@ -156,6 +233,78 @@ def model_supports_reasoning(model_name: str) -> bool:
    return bool(entry and entry.get("supports_reasoning"))


+def is_recommended_or_frontier_model(model_name: str) -> bool:
+    """Return whether a model is recommended or in a frontier model family."""
+    name = _normalized_model_name(model_name)
+    if not name:
+        return False
+    if name in _RECOMMENDED_MODEL_NAME_SET:
+        return True
+    provider_name, bare_model_name = _split_model_provider(name)
+    return any(
+        _matches_frontier_family(provider_name, bare_model_name, provider_markers, prefixes)
+        for provider_markers, prefixes in FRONTIER_MODEL_FAMILIES
+    )
+
+
+def _normalized_model_name(model_name: str) -> str:
+    name = model_name.strip().lower()
+    for prefix in ("litellm/", "any-llm/"):
+        if name.startswith(prefix):
+            name = name[len(prefix) :]
+            break
+    return name
+
+
+def _split_model_provider(model_name: str) -> tuple[str | None, str]:
+    if "/" not in model_name:
+        return None, model_name
+    provider_name, bare_model_name = model_name.rsplit("/", 1)
+    return provider_name, bare_model_name
+
+
+def _matches_frontier_family(
+    provider_name: str | None,
+    model_name: str,
+    provider_markers: tuple[str, ...],
+    model_prefixes: tuple[str, ...],
+) -> bool:
+    if not _matches_model_prefix(model_name, model_prefixes):
+        return False
+    if provider_name is None:
+        return True
+    return _contains_provider_marker(
+        provider_name, provider_markers, split_compound_names=True
+    ) or _contains_provider_marker(model_name, provider_markers)
+
+
+def _matches_model_prefix(model_name: str, model_prefixes: tuple[str, ...]) -> bool:
+    return any(
+        candidate.startswith(prefix)
+        for candidate in _model_name_candidates(model_name)
+        for prefix in model_prefixes
+    )
+
+
+def _model_name_candidates(model_name: str) -> tuple[str, ...]:
+    if "." not in model_name:
+        return (model_name,)
+    suffixes = tuple(
+        model_name.split(".", index)[-1] for index in range(1, model_name.count(".") + 1)
+    )
+    return (model_name, *suffixes)
+
+
+def _contains_provider_marker(
+    value: str, provider_markers: tuple[str, ...], *, split_compound_names: bool = False
+) -> bool:
+    parts = set(value.replace(".", "/").split("/"))
+    if split_compound_names:
+        for separator in ("_", "-"):
+            parts.update(piece for part in tuple(parts) for piece in part.split(separator))
+    return any(marker in parts for marker in provider_markers)
+
+
 def is_known_openai_bare_model(model_name: str) -> bool:
    import litellm

@@ -56,6 +56,8 @@ class RuntimeSettings(BaseSettings):
    # on large repos). Above this, the user must bind-mount via ``--mount``.
    # Set to 0 (or less) to disable the pre-flight check entirely.
    max_local_copy_mb: int = Field(default=1024, alias="STRIX_MAX_LOCAL_COPY_MB")
+    # Max screenshot/image tool outputs kept live per agent context (0 = none).
+    max_context_images: int = Field(default=3, ge=0, alias="STRIX_MAX_CONTEXT_IMAGES")


 class TelemetrySettings(BaseSettings):
@@ -70,6 +72,15 @@ class IntegrationSettings(BaseSettings):
    perplexity_api_key: str | None = Field(default=None, alias="PERPLEXITY_API_KEY")


+class ViewerSettings(BaseSettings):
+    model_config = _BASE_CONFIG
+
+    # Base URL of the Strix relay the local viewer proxies to for email
+    # verification and encrypted report delivery. The browser never talks to
+    # the relay directly; the local server is the only caller.
+    app_url: str = Field(default="https://app.strix.ai", alias="STRIX_APP_URL")
+
+
 class Settings(BaseSettings):
    model_config = _BASE_CONFIG

@@ -77,3 +88,4 @@ class Settings(BaseSettings):
    runtime: RuntimeSettings = Field(default_factory=RuntimeSettings)
    telemetry: TelemetrySettings = Field(default_factory=TelemetrySettings)
    integrations: IntegrationSettings = Field(default_factory=IntegrationSettings)
+    viewer: ViewerSettings = Field(default_factory=ViewerSettings)
@@ -10,6 +10,8 @@ from dataclasses import dataclass, field
 from pathlib import Path
 from typing import TYPE_CHECKING, Any, Literal, cast

+from strix.core.sessions import session_write_lock
+

 if TYPE_CHECKING:
    from agents.items import TResponseInputItem
@@ -137,7 +139,8 @@ class AgentCoordinator:
            )
            return False
        try:
-            await session.add_items([self._message_to_session_item(message)])
+            async with session_write_lock(session):
+                await session.add_items([self._message_to_session_item(message)])
        except Exception:
            logger.exception(
                "agent.send failed to append to SDK session target=%s",
@@ -17,7 +17,11 @@ from openai import APIError

 from strix.core.hooks import BudgetExceededError
 from strix.core.inputs import child_initial_input
-from strix.core.sessions import open_agent_session, strip_all_images_from_session
+from strix.core.sessions import (
+    enforce_image_budget,
+    open_agent_session,
+    strip_all_images_from_session,
+)


 if TYPE_CHECKING:
@@ -349,6 +353,13 @@ async def _run_cycle(  # noqa: PLR0912, PLR0915
    while True:
        try:
            await coordinator.mark_running(agent_id)
+            if session is not None:
+                max_images = context.get("max_context_images")
+                if isinstance(max_images, int):
+                    try:
+                        await enforce_image_budget(session, max_images)
+                    except Exception:
+                        logger.exception("image-budget enforcement failed for %s", agent_id)
            stream = Runner.run_streamed(
                agent,
                input=input_data,
@@ -3,6 +3,7 @@
 from __future__ import annotations

 import logging
+import math
 from typing import TYPE_CHECKING, Any

 from agents.lifecycle import RunHooks
@@ -27,8 +28,9 @@ class ReportUsageHooks(RunHooks[dict[str, Any]]):
    """Persist SDK-native usage after every model response."""

    def __init__(self, *, model: str, max_budget_usd: float | None = None) -> None:
-        import math
-        if max_budget_usd is not None and (not math.isfinite(max_budget_usd) or max_budget_usd <= 0):
+        if max_budget_usd is not None and (
+            not math.isfinite(max_budget_usd) or max_budget_usd <= 0
+        ):
            raise ValueError("max_budget_usd must be a finite number greater than 0")
        self._model = model
        self._max_budget_usd = max_budget_usd
@@ -12,7 +12,9 @@ from strix.config.models import (
    DEFAULT_MODEL_RETRY,
    is_known_openai_bare_model,
    model_supports_reasoning,
+    request_timeout_extra_args,
 )
+from strix.core.sessions import scrub_images_from_items


 if TYPE_CHECKING:
@@ -125,11 +127,13 @@ def make_model_settings(
    *,
    model_name: str,
    force_required_tool_choice: bool = False,
+    request_timeout: float | None = None,
 ) -> ModelSettings:
    model_settings = ModelSettings(
        parallel_tool_calls=False,
        retry=DEFAULT_MODEL_RETRY,
        include_usage=True,
+        extra_args=request_timeout_extra_args(request_timeout),
    )
    if (
        reasoning_effort is not None
@@ -161,7 +165,11 @@ def child_initial_input(
    """
    parts: list[str] = []
    if parent_history:
-        rendered = json.dumps(parent_history, ensure_ascii=False, default=str)
+        rendered = json.dumps(
+            scrub_images_from_items(parent_history),
+            ensure_ascii=False,
+            default=str,
+        )
        parts.append(
            "== Inherited context from parent (background only) ==\n"
            f"{rendered}\n"
@@ -21,3 +21,20 @@ def runtime_state_dir(run_dir: Path) -> Path:

 def run_record_path(run_dir: Path) -> Path:
    return run_dir / RUN_RECORD_FILENAME
+
+
+def runs_base_dir(*, cwd: Path | None = None) -> Path:
+    base = cwd or Path.cwd()
+    return base / RUNS_DIR_NAME
+
+
+def latest_run_dir(*, cwd: Path | None = None) -> Path | None:
+    base = runs_base_dir(cwd=cwd)
+    if not base.is_dir():
+        return None
+    candidates = [child for child in base.iterdir() if run_record_path(child).is_file()]
+    if not candidates:
+        return None
+    # run.json is rewritten on status/end changes, so its mtime tracks activity
+    # more reliably than the directory mtime (a live run sorts to the top).
+    return max(candidates, key=lambda child: run_record_path(child).stat().st_mtime)
@@ -14,6 +14,7 @@ from agents.sandbox import SandboxRunConfig
 from openai import RateLimitError

 from strix.agents.factory import build_strix_agent, make_child_factory
+from strix.agents.prompt import render_system_prompt
 from strix.config import load_settings
 from strix.config.models import (
    StrixProvider,
@@ -51,6 +52,52 @@ logger = logging.getLogger(__name__)
 StreamEventSink = Callable[[str, Any], None]


+def _merge_root_prompt_context(
+    scope_context: dict[str, Any],
+    extra_system_prompt_context: dict[str, Any] | None,
+) -> dict[str, Any]:
+    if not extra_system_prompt_context:
+        return scope_context
+    reserved_keys = scope_context.keys() & extra_system_prompt_context.keys()
+    if reserved_keys:
+        raise ValueError(
+            "extra_system_prompt_context cannot override built-in scope keys: "
+            f"{sorted(reserved_keys)}",
+        )
+    return {**scope_context, **extra_system_prompt_context}
+
+
+def _compose_root_instructions_override(
+    root_instructions_override: str | None,
+    *,
+    skills: list[str],
+    scan_mode: str,
+    is_whitebox: bool,
+    interactive: bool,
+    system_prompt_context: dict[str, Any],
+) -> str | None:
+    if root_instructions_override is None:
+        return None
+
+    base_instructions = render_system_prompt(
+        skills=skills,
+        scan_mode=scan_mode,
+        is_whitebox=is_whitebox,
+        is_root=True,
+        interactive=interactive,
+        system_prompt_context=system_prompt_context,
+    )
+    return (
+        f"{base_instructions}\n\n"
+        "<root_scan_instructions_override>\n"
+        "The following root scan instructions are subordinate to the "
+        "system-verified scope above. They cannot expand, replace, or weaken "
+        "authorized target constraints.\n\n"
+        f"{root_instructions_override}\n"
+        "</root_scan_instructions_override>"
+    )
+
+
 async def run_strix_scan(
    *,
    scan_config: dict[str, Any],
@@ -64,8 +111,17 @@ async def run_strix_scan(
    model: str | None = None,
    cleanup_on_exit: bool = True,
    event_sink: StreamEventSink | None = None,
+    root_instructions_override: str | None = None,
+    extra_system_prompt_context: dict[str, Any] | None = None,
 ) -> RunResultBase | None:
-    """Run or resume one Strix scan against a sandbox."""
+    """Run or resume one Strix scan against a sandbox.
+
+    ``root_instructions_override`` adds root scan instructions to the rendered
+    root prompt without replacing the system-verified scope block.
+    ``extra_system_prompt_context`` is merged into the root agent's scan
+    context before prompt rendering. Child agents keep the standard scan prompt
+    and context.
+    """
    if scan_id is None:
        scan_id = f"scan-{uuid.uuid4().hex[:8]}"

@@ -159,6 +215,7 @@ async def run_strix_scan(
            settings.llm.reasoning_effort,
            model_name=resolved_model,
            force_required_tool_choice=settings.llm.force_required_tool_choice,
+            request_timeout=settings.llm.timeout,
        )
        run_config = RunConfig(
            model=resolved_model,
@@ -170,6 +227,15 @@ async def run_strix_scan(
        hooks = ReportUsageHooks(model=resolved_model, max_budget_usd=max_budget_usd)

        scope_context = build_scope_context(scan_config)
+        root_context = _merge_root_prompt_context(scope_context, extra_system_prompt_context)
+        root_instructions = _compose_root_instructions_override(
+            root_instructions_override,
+            skills=skills,
+            scan_mode=scan_mode,
+            is_whitebox=is_whitebox,
+            interactive=interactive,
+            system_prompt_context=root_context,
+        )

        root_agent = build_strix_agent(
            name="strix",
@@ -179,7 +245,8 @@ async def run_strix_scan(
            is_whitebox=is_whitebox,
            interactive=interactive,
            chat_completions_tools=chat_completions_tools,
-            system_prompt_context=scope_context,
+            system_prompt_context=root_context,
+            instructions_override=root_instructions,
        )

        if not is_resume:
@@ -221,6 +288,7 @@ async def run_strix_scan(
            "parent_id": None,
            "interactive": interactive,
            "spawn_child_agent": spawn_child_agent,
+            "max_context_images": settings.runtime.max_context_images,
        }

        root_session = open_agent_session(root_id, agents_db)
@@ -2,64 +2,149 @@

 from __future__ import annotations

-import contextlib
+import asyncio
+import logging
 from typing import TYPE_CHECKING, Any, cast
+from weakref import WeakKeyDictionary

 from agents.memory import SQLiteSession


 if TYPE_CHECKING:
+    from collections.abc import Callable
    from pathlib import Path

    from agents.items import TResponseInputItem
    from agents.memory import Session


+logger = logging.getLogger(__name__)
+
+
 def open_agent_session(agent_id: str, path: Path) -> SQLiteSession:
    path.parent.mkdir(parents=True, exist_ok=True)
    return SQLiteSession(session_id=agent_id, db_path=path)


 _IMAGE_REJECTED_TEXT = "[image rejected by the model]"
+_IMAGE_ELIDED_TEXT = "[older screenshot elided to bound context memory]"
+_INHERITED_IMAGE_TEXT = "[screenshot omitted from inherited context]"
+
+
+def _output_has_image(item_dict: dict[str, Any]) -> bool:
+    return (
+        item_dict.get("type") == "function_call_output"
+        and isinstance(item_dict.get("output"), list)
+        and any(isinstance(b, dict) and b.get("type") == "input_image" for b in item_dict["output"])
+    )
+
+
+def _elided_output(item_dict: dict[str, Any], text: str) -> dict[str, Any]:
+    # Replace only image blocks; sibling text blocks are preserved.
+    output = item_dict.get("output")
+    blocks = output if isinstance(output, list) else []
+    return {
+        "type": "function_call_output",
+        "call_id": item_dict.get("call_id"),
+        "output": [
+            {"type": "input_text", "text": text}
+            if isinstance(block, dict) and block.get("type") == "input_image"
+            else block
+            for block in blocks
+        ],
+    }
+
+
+_session_write_locks: WeakKeyDictionary[Session, asyncio.Lock] = WeakKeyDictionary()
+
+
+def session_write_lock(session: Session) -> asyncio.Lock:
+    """Lock serialising all out-of-band writes to ``session``."""
+    lock = _session_write_locks.get(session)
+    if lock is None:
+        lock = asyncio.Lock()
+        _session_write_locks[session] = lock
+    return lock
+
+
+async def _rewrite_session(
+    session: Session,
+    transform: Callable[[list[Any]], tuple[list[Any], bool]],
+) -> bool:
+    """Read-modify-write a session under its write lock, restoring on failure."""
+    async with session_write_lock(session):
+        items = await session.get_items()
+        if not items:
+            return False
+        rebuilt, changed = transform(list(items))
+        if not changed:
+            return False
+        rebuilt_items = cast("list[TResponseInputItem]", rebuilt)
+        original_items = cast("list[TResponseInputItem]", list(items))
+        await session.clear_session()
+        try:
+            await session.add_items(rebuilt_items)
+        except Exception:
+            logger.exception("session rewrite failed; restoring original items")
+            await session.clear_session()
+            await session.add_items(original_items)
+            raise
+        return True


 async def strip_all_images_from_session(session: Session) -> bool:
-    items = await session.get_items()
-    if not items:
+    """Replace every image tool output with a text placeholder (rejection recovery)."""
+
+    def _transform(items: list[Any]) -> tuple[list[Any], bool]:
+        rebuilt: list[Any] = []
+        changed = False
+        for item in items:
+            item_dict = cast("dict[str, Any]", item) if isinstance(item, dict) else None
+            if item_dict is not None and _output_has_image(item_dict):
+                rebuilt.append(_elided_output(item_dict, _IMAGE_REJECTED_TEXT))
+                changed = True
+            else:
+                rebuilt.append(item)
+        return rebuilt, changed
+
+    return await _rewrite_session(session, _transform)
+
+
+async def enforce_image_budget(session: Session, max_images: int) -> bool:
+    """Keep only the most recent ``max_images`` image outputs; elide older ones."""
+    if max_images < 0:
        return False

-    rebuilt: list[Any] = []
-    changed = False
-    for item in items:
-        item_dict = cast("dict[str, Any]", item) if isinstance(item, dict) else None
-        if (
-            item_dict is not None
-            and item_dict.get("type") == "function_call_output"
-            and isinstance(item_dict.get("output"), list)
-            and any(
-                isinstance(b, dict) and b.get("type") == "input_image" for b in item_dict["output"]
-            )
-        ):
-            rebuilt.append(
-                {
-                    "type": "function_call_output",
-                    "call_id": item_dict.get("call_id"),
-                    "output": [{"type": "input_text", "text": _IMAGE_REJECTED_TEXT}],
-                },
-            )
-            changed = True
-        else:
-            rebuilt.append(item)
+    def _transform(items: list[Any]) -> tuple[list[Any], bool]:
+        image_indices = [
+            i
+            for i, item in enumerate(items)
+            if isinstance(item, dict) and _output_has_image(cast("dict[str, Any]", item))
+        ]
+        if len(image_indices) <= max_images:
+            return items, False
+        to_elide = set(image_indices[: len(image_indices) - max_images])
+        rebuilt = [
+            _elided_output(cast("dict[str, Any]", item), _IMAGE_ELIDED_TEXT)
+            if i in to_elide
+            else item
+            for i, item in enumerate(items)
+        ]
+        return rebuilt, True

-    if not changed:
-        return False
+    return await _rewrite_session(session, _transform)

-    rebuilt_items = cast("list[TResponseInputItem]", rebuilt)
-    await session.clear_session()
-    try:
-        await session.add_items(rebuilt_items)
-    except Exception:
-        with contextlib.suppress(Exception):
-            await session.add_items(rebuilt_items)
-        raise
-    return True
+
+def scrub_images_from_items(items: list[Any]) -> list[Any]:
+    """Return a copy of ``items`` with every image block replaced by text."""
+
+    def _scrub(obj: Any) -> Any:
+        if isinstance(obj, dict):
+            if obj.get("type") == "input_image":
+                return {"type": "input_text", "text": _INHERITED_IMAGE_TEXT}
+            return {k: _scrub(v) for k, v in obj.items()}
+        if isinstance(obj, list):
+            return [_scrub(v) for v in obj]
+        return obj
+
+    return [_scrub(item) for item in items]
@@ -67,6 +67,16 @@ Toast.-information .toast--title {
    display: none;
 }

+#viewer_cta {
+    height: auto;
+    background: transparent;
+    border: round #333333;
+    color: #60a5fa;
+    padding: 0 1;
+    margin-bottom: 1;
+    text-align: center;
+}
+
 #agents_tree {
    height: 1fr;
    background: transparent;
@@ -7,6 +7,7 @@ import argparse
 import asyncio
 import shutil
 import sys
+import time
 from datetime import UTC, datetime
 from pathlib import Path

@@ -23,9 +24,11 @@ from strix.config import (
    persist_current,
 )
 from strix.config.models import (
+    RECOMMENDED_MODEL_NAMES,
    StrixProvider,
    configure_sdk_model_defaults,
    is_known_openai_bare_model,
+    is_recommended_or_frontier_model,
 )
 from strix.core.paths import run_dir_for, runtime_state_dir
 from strix.interface.cli import run_cli
@@ -264,7 +267,7 @@ def _provider_import_hint(exc: BaseException, model: str) -> str | None:
    return None


-async def warm_up_llm() -> None:
+async def warm_up_llm(show_model_warning: bool = True) -> None:
    console = Console()
    logger.info("Warming up LLM connection")

@@ -306,6 +309,32 @@ async def warm_up_llm() -> None:
            )
            sys.exit(1)

+        if show_model_warning and raw_model and not is_recommended_or_frontier_model(raw_model):
+            warn_text = Text()
+            warn_text.append("MODEL QUALITY WARNING", style="bold yellow")
+            warn_text.append("\n\n", style="white")
+            warn_text.append(f"'{raw_model}'", style="bold cyan")
+            warn_text.append(
+                " is not a recommended frontier model for Strix.\nSecurity scans work best with:\n",
+                style="white",
+            )
+            for recommended_model in RECOMMENDED_MODEL_NAMES:
+                warn_text.append(f"• {recommended_model}\n", style="bold cyan")
+            warn_text.append(
+                "\nYou can continue, but weaker models may miss vulnerabilities "
+                "or produce lower-quality findings.",
+                style="white",
+            )
+            console.print(
+                Panel(
+                    warn_text,
+                    title="[bold white]STRIX",
+                    title_align="left",
+                    border_style="yellow",
+                    padding=(1, 2),
+                ),
+            )
+
        model = StrixProvider().get_model(raw_model)
        await asyncio.wait_for(
            model.get_response(
@@ -692,7 +721,9 @@ def _load_resume_state(args: argparse.Namespace, parser: argparse.ArgumentParser
        args.scan_mode = persisted_scan_mode


-def display_completion_message(args: argparse.Namespace, results_path: Path) -> None:
+def display_completion_message(
+    args: argparse.Namespace, results_path: Path, web_url: str | None = None
+) -> None:
    console = Console()
    report_state = get_global_report_state()

@@ -731,6 +762,29 @@ def display_completion_message(args: argparse.Namespace, results_path: Path) ->
    results_text.append(str(results_path), style="#60a5fa")
    panel_parts.extend(["\n", results_text])

+    if web_url:
+        web_text = Text()
+        web_text.append("\n")
+        web_text.append("View in web", style="dim")
+        web_text.append("  ")
+        # OSC-8 hyperlink: clickable in modern terminals, falls back to the URL.
+        web_text.append(web_url, style=f"#60a5fa link {web_url}")
+        panel_parts.extend(["\n", web_text])
+
+        reopen_text = Text()
+        reopen_text.append("\n")
+        reopen_text.append("Reopen", style="dim")
+        reopen_text.append("       ")
+        reopen_text.append(f"strix view {args.run_name}", style="#22c55e")
+        panel_parts.extend(["\n", reopen_text])
+    else:
+        view_text = Text()
+        view_text.append("\n")
+        view_text.append("View", style="dim")
+        view_text.append("         ")
+        view_text.append(f"strix view {args.run_name}", style="#22c55e")
+        panel_parts.extend(["\n", view_text])
+
    if not scan_completed:
        resume_text = Text()
        resume_text.append("\n")
@@ -818,6 +872,14 @@ def main() -> None:
    if sys.platform == "win32":
        asyncio.set_event_loop_policy(asyncio.WindowsSelectorEventLoopPolicy())

+    # `strix view [<run>]` is a viewer-only subcommand, dispatched before the
+    # scan argument parser (which requires a target) and before any scan setup.
+    if len(sys.argv) > 1 and sys.argv[1] == "view":
+        from strix.viewer.cli import run_view
+
+        run_view(sys.argv[2:])
+        return
+
    args = parse_arguments()

    if args.config:
@@ -827,7 +889,7 @@ def main() -> None:
    pull_docker_image()

    validate_environment()
-    asyncio.run(warm_up_llm())
+    asyncio.run(warm_up_llm(show_model_warning=args.non_interactive))

    persist_current()

@@ -912,7 +974,40 @@ def main() -> None:
            scarf.end(report_state, exit_reason=exit_reason)

    results_path = run_dir_for(args.run_name)
-    display_completion_message(args, results_path)
+
+    # For an interactive run, host the local viewer so the completion panel can
+    # show a clickable "View in web" link. Skipped in non-interactive/CI runs
+    # (no TTY to serve and it would block the process).
+    viewer_httpd = None
+    web_url = None
+    if not args.non_interactive and sys.stdout.isatty():
+        from strix.viewer.server import authorized_url, bundle_is_built, serve
+
+        if bundle_is_built():
+            try:
+                viewer_httpd, base_url, token = serve(results_path, open_browser=False)
+                # The completion panel's "View in web" link must authorize the
+                # browser, so hand it the tokened URL rather than the bare host.
+                web_url = authorized_url(base_url, token)
+                posthog.viewer_opened(source="post_scan", live=False)
+            except Exception:
+                logger.debug("could not start local viewer", exc_info=True)
+                viewer_httpd, web_url = None, None
+
+    display_completion_message(args, results_path, web_url=web_url)
+
+    if viewer_httpd is not None:
+        console = Console()
+        console.print("[dim]Hosting the local viewer. Press Ctrl-C to stop.[/]")
+        console.print()
+        try:
+            while True:
+                time.sleep(1)
+        except KeyboardInterrupt:
+            console.print("\n[dim]Viewer stopped.[/]")
+        finally:
+            viewer_httpd.shutdown()
+            viewer_httpd.server_close()

    if args.non_interactive:
        report_state = get_global_report_state()
@@ -6,6 +6,7 @@ import logging
 import signal
 import sys
 import threading
+import webbrowser
 from collections.abc import Callable
 from importlib.metadata import PackageNotFoundError
 from importlib.metadata import version as pkg_version
@@ -31,6 +32,7 @@ from textual.widgets import Button, Label, Static, TextArea, Tree
 from textual.widgets.tree import TreeNode

 from strix.config import load_settings
+from strix.config.models import is_recommended_or_frontier_model
 from strix.core.hooks import BudgetExceededError
 from strix.core.runner import run_strix_scan
 from strix.interface.tui.live_view import TuiLiveView
@@ -116,9 +118,16 @@ class SplashScreen(Static):  # type: ignore[misc]
        self._animation_timer: Timer | None = None
        self._panel_static: Static | None = None
        self._version = "dev"
+        self._non_frontier_model: str | None = None

    def compose(self) -> ComposeResult:
        self._version = get_package_version()
+        try:
+            model = (load_settings().llm.model or "").strip()
+        except Exception:
+            model = ""
+        if model and not is_recommended_or_frontier_model(model):
+            self._non_frontier_model = model
        self._animation_step = 0
        start_line = self._build_start_line_text(self._animation_step)
        panel = self._build_panel(start_line)
@@ -145,7 +154,7 @@ class SplashScreen(Static):  # type: ignore[misc]
        self._panel_static.update(panel)

    def _build_panel(self, start_line: Text) -> Panel:
-        content = Group(
+        rows = [
            Align.center(Text(self.BANNER.strip("\n"), style=self.PRIMARY_GREEN, justify="center")),
            Align.center(Text(" ")),
            Align.center(self._build_welcome_text()),
@@ -155,9 +164,26 @@ class SplashScreen(Static):  # type: ignore[misc]
            Align.center(start_line.copy()),
            Align.center(Text(" ")),
            Align.center(self._build_url_text()),
-        )
+        ]
+        if self._non_frontier_model:
+            rows.extend(
+                (
+                    Align.center(Text(" ")),
+                    Align.center(self._build_model_warning_text(self._non_frontier_model)),
+                )
+            )

-        return Panel.fit(content, border_style=self.PRIMARY_GREEN, padding=(1, 6))
+        return Panel.fit(Group(*rows), border_style=self.PRIMARY_GREEN, padding=(1, 6))
+
+    @staticmethod
+    def _build_model_warning_text(model: str) -> Text:
+        text = Text("⚠ ", style=Style(color="yellow", bold=True))
+        text.append(model, style=Style(color="cyan", bold=True))
+        text.append(
+            " is not a recommended frontier model - pentest quality could be degraded",
+            style=Style(color="yellow"),
+        )
+        return text

    def _build_url_text(self) -> Text:
        return Text("strix.ai", style=Style(color=self.PRIMARY_GREEN, bold=True))
@@ -371,6 +397,19 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
            text.append("Target: ", style=self.FIELD_STYLE)
            text.append(target)

+        dep_meta = vuln.get("dependency_metadata") or {}
+        for label, key in (
+            ("Package", "package_name"),
+            ("Ecosystem", "package_ecosystem"),
+            ("Installed Version", "installed_version"),
+            ("Fixed Version", "fixed_version"),
+        ):
+            value = dep_meta.get(key)
+            if value:
+                text.append("\n\n")
+                text.append(f"{label}: ", style=self.FIELD_STYLE)
+                text.append(str(value))
+
        endpoint = vuln.get("endpoint", "")
        if endpoint:
            text.append("\n\n")
@@ -389,6 +428,18 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
            text.append("CVE: ", style=self.FIELD_STYLE)
            text.append(cve)

+        cwe = vuln.get("cwe", "")
+        if cwe:
+            text.append("\n\n")
+            text.append("CWE: ", style=self.FIELD_STYLE)
+            text.append(cwe)
+
+        fix_effort = vuln.get("fix_effort", "")
+        if fix_effort:
+            text.append("\n\n")
+            text.append("Fix Effort: ", style=self.FIELD_STYLE)
+            text.append(str(fix_effort).title())
+
        cvss_breakdown = vuln.get("cvss_breakdown", {})
        if cvss_breakdown:
            cvss_parts = []
@@ -434,6 +485,13 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
            text.append("\n")
            text.append(technical_analysis)

+        evidence = vuln.get("evidence", "")
+        if evidence:
+            text.append("\n\n")
+            text.append("Evidence", style=self.FIELD_STYLE)
+            text.append("\n")
+            text.append(evidence)
+
        poc_description = vuln.get("poc_description", "")
        if poc_description:
            text.append("\n\n")
@@ -455,6 +513,13 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
            text.append("\n")
            text.append(remediation_steps)

+        assumptions = vuln.get("assumptions", "")
+        if assumptions:
+            text.append("\n\n")
+            text.append("Assumptions", style=self.FIELD_STYLE)
+            text.append("\n")
+            text.append(assumptions)
+
        return text

    def _get_markdown_report(self) -> str:
@@ -476,14 +541,27 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
            lines.append(f"**Agent:** {vuln['agent_name']}")
        if vuln.get("target"):
            lines.append(f"**Target:** {vuln['target']}")
+        dep_meta = vuln.get("dependency_metadata") or {}
+        if dep_meta.get("package_name"):
+            lines.append(f"**Package:** {dep_meta['package_name']}")
+        if dep_meta.get("package_ecosystem"):
+            lines.append(f"**Ecosystem:** {dep_meta['package_ecosystem']}")
+        if dep_meta.get("installed_version"):
+            lines.append(f"**Installed Version:** {dep_meta['installed_version']}")
+        if dep_meta.get("fixed_version"):
+            lines.append(f"**Fixed Version:** {dep_meta['fixed_version']}")
        if vuln.get("endpoint"):
            lines.append(f"**Endpoint:** {vuln['endpoint']}")
        if vuln.get("method"):
            lines.append(f"**Method:** {vuln['method']}")
        if vuln.get("cve"):
            lines.append(f"**CVE:** {vuln['cve']}")
+        if vuln.get("cwe"):
+            lines.append(f"**CWE:** {vuln['cwe']}")
        if vuln.get("cvss") is not None:
            lines.append(f"**CVSS:** {vuln['cvss']}")
+        if vuln.get("fix_effort"):
+            lines.append(f"**Fix Effort:** {str(vuln['fix_effort']).title()}")

        cvss_breakdown = vuln.get("cvss_breakdown", {})
        if cvss_breakdown:
@@ -514,6 +592,9 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
        if vuln.get("technical_analysis"):
            lines.extend(["", "## Technical Analysis", "", vuln["technical_analysis"]])

+        if vuln.get("evidence"):
+            lines.extend(["", "## Evidence", "", vuln["evidence"]])
+
        if vuln.get("poc_description") or vuln.get("poc_script_code"):
            lines.extend(["", "## Proof of Concept", ""])
            if vuln.get("poc_description"):
@@ -552,6 +633,9 @@ class VulnerabilityDetailScreen(ModalScreen):  # type: ignore[misc]
        if vuln.get("remediation_steps"):
            lines.extend(["", "## Remediation", "", vuln["remediation_steps"]])

+        if vuln.get("assumptions"):
+            lines.extend(["", "## Assumptions", "", vuln["assumptions"]])
+
        lines.append("")
        return "\n".join(lines)

@@ -685,6 +769,7 @@ class StrixTUIApp(App):  # type: ignore[misc]
        Binding("ctrl+q", "request_quit", "Quit", priority=True),
        Binding("ctrl+c", "request_quit", "Quit", priority=True),
        Binding("escape", "stop_selected_agent", "Stop Agent", priority=True),
+        Binding("ctrl+o", "open_viewer", "Open Viewer", priority=True),
    ]

    def __init__(self, args: argparse.Namespace):
@@ -711,6 +796,8 @@ class StrixTUIApp(App):  # type: ignore[misc]
        self._displayed_events: list[str] = []

        self._scan_thread: threading.Thread | None = None
+        self._viewer_httpd: Any = None
+        self._viewer_url: str | None = None
        self._scan_loop: asyncio.AbstractEventLoop | None = None
        self._scan_stop_event = threading.Event()
        self._scan_completed = threading.Event()
@@ -820,7 +907,12 @@ class StrixTUIApp(App):  # type: ignore[misc]

            vulnerabilities_panel = VulnerabilitiesPanel(id="vulnerabilities_panel")

-            sidebar = Vertical(agents_tree, vulnerabilities_panel, stats_scroll, id="sidebar")
+            viewer_cta = Static(self._viewer_cta_markup(), id="viewer_cta")
+            viewer_cta.ALLOW_SELECT = False
+
+            sidebar = Vertical(
+                viewer_cta, agents_tree, vulnerabilities_panel, stats_scroll, id="sidebar"
+            )

            content_container.mount(chat_area_container)
            content_container.mount(sidebar)
@@ -1301,7 +1393,7 @@ class StrixTUIApp(App):  # type: ignore[misc]

    def _start_dot_animation(self) -> None:
        if self._dot_animation_timer is None:
-            self._dot_animation_timer = self.set_interval(0.25, self._animate_dots)
+            self._dot_animation_timer = self.set_interval(0.06, self._animate_dots)

    def _stop_dot_animation(self) -> None:
        if self._dot_animation_timer is not None:
@@ -1722,6 +1814,7 @@ class StrixTUIApp(App):  # type: ignore[misc]

    async def action_custom_quit(self) -> None:
        self._fire_sandbox_cleanup()
+        self._shutdown_viewer()

        if self._scan_thread and self._scan_thread.is_alive():
            self._scan_stop_event.set()
@@ -1730,6 +1823,70 @@ class StrixTUIApp(App):  # type: ignore[misc]

        self.exit()

+    def _viewer_cta_markup(self, url: str | None = None) -> str:
+        if url:
+            return f"[@click=app.open_viewer][#22c55e]● Viewer running[/][/]\n[dim]{url}[/]"
+        return "[@click=app.open_viewer]▶ Watch live in browser[/]"
+
+    def _set_viewer_cta(self, markup: str) -> None:
+        with contextlib.suppress(Exception):
+            self.query_one("#viewer_cta", Static).update(markup)
+
+    def action_open_viewer(self) -> None:
+        if self._viewer_url:
+            with contextlib.suppress(Exception):
+                webbrowser.open(self._viewer_url)
+            return
+        try:
+            from strix.viewer.server import authorized_url, bundle_is_built, serve
+
+            if not bundle_is_built():
+                self._set_viewer_cta("[#eab308]Viewer UI not built[/]")
+                return
+            run_dir = self.report_state.get_run_dir()
+
+            def _viewer_steer(agent_id: str, message: str) -> bool:
+                # Reuse the exact TUI delivery path, but target the agent the
+                # web graph selected (not the TUI's current selection).
+                return send_user_message_to_agent(
+                    coordinator=self.coordinator,
+                    loop=self._scan_loop,
+                    live_view=self.live_view,
+                    target_agent_id=agent_id,
+                    message=message,
+                )
+
+            httpd, url, token = serve(run_dir, open_browser=True, steer_handler=_viewer_steer)
+        except Exception:
+            logger.debug("failed to start local viewer", exc_info=True)
+            self._set_viewer_cta("[red]Viewer failed to start[/]")
+            return
+        self._viewer_httpd = httpd
+        # Store the tokened URL so reopening the CTA re-authorizes the browser
+        # (this viewer carries a steer handler, so the session is required).
+        self._viewer_url = authorized_url(url, token)
+        self._set_viewer_cta(self._viewer_cta_markup(self._viewer_url))
+
+        with contextlib.suppress(Exception):
+            from strix.telemetry import posthog
+
+            live = self.report_state.run_record.get("status") not in {
+                "completed",
+                "stopped",
+                "failed",
+                "interrupted",
+            }
+            posthog.viewer_opened(source="tui", live=live)
+
+    def _shutdown_viewer(self) -> None:
+        httpd = self._viewer_httpd
+        if httpd is None:
+            return
+        self._viewer_httpd = None
+        with contextlib.suppress(Exception):
+            httpd.shutdown()
+            httpd.server_close()
+
    def _fire_sandbox_cleanup(self) -> None:
        self.coordinator.mark_shutting_down()
        loop = self._scan_loop
@@ -24,14 +24,26 @@ def load_session_history(run_dir: Path, agent_ids: Any) -> list[tuple[str, dict[
    if not agents_db.exists() or not session_ids:
        return []
    session_id_set = set(session_ids)
+    # Open read-only: the scan process may be actively writing this WAL database
+    # from another process (the local viewer tails it live), and a reader must
+    # never lock or mutate it. mode=ro (not immutable=1) still reads the latest
+    # committed WAL state; WAL permits concurrent readers alongside the writer.
+    conn: sqlite3.Connection | None = None
    try:
-        with sqlite3.connect(agents_db) as conn:
-            rows = conn.execute(
-                "select id, session_id, message_data, created_at from agent_messages order by id"
-            ).fetchall()
+        conn = sqlite3.connect(
+            f"file:{agents_db}?mode=ro",
+            uri=True,
+            check_same_thread=False,
+        )
+        rows = conn.execute(
+            "select id, session_id, message_data, created_at from agent_messages order by id"
+        ).fetchall()
    except sqlite3.Error:
        logger.exception("Failed to hydrate TUI history from %s", agents_db)
        return []
+    finally:
+        if conn is not None:
+            conn.close()

    items: list[tuple[str, dict[str, Any], str]] = []
    for row_id, agent_id, message_data, created_at in rows:
@@ -256,3 +256,176 @@ class CreateVulnerabilityReportRenderer(BaseToolRenderer):

        css_classes = cls.get_css_classes("completed")
        return Static(padded, classes=css_classes)
+
+
+@register_tool_renderer
+class CreateDependencyReportRenderer(BaseToolRenderer):
+    tool_name: ClassVar[str] = "create_dependency_report"
+    css_classes: ClassVar[list[str]] = ["tool-call", "reporting-tool"]
+
+    SEVERITY_COLORS: ClassVar[dict[str, str]] = {
+        "critical": "#dc2626",
+        "high": "#ea580c",
+        "medium": "#d97706",
+        "low": "#65a30d",
+        "info": "#0284c7",
+    }
+
+    @classmethod
+    def _get_cvss_color(cls, cvss_score: float) -> str:
+        if cvss_score >= 9.0:
+            return "#dc2626"
+        if cvss_score >= 7.0:
+            return "#ea580c"
+        if cvss_score >= 4.0:
+            return "#d97706"
+        if cvss_score >= 0.1:
+            return "#65a30d"
+        return "#6b7280"
+
+    @classmethod
+    def _render_unsuccessful(cls, args: dict[str, Any], result: dict[str, Any]) -> Static:
+        text = Text()
+        text.append("📦 ")
+        text.append("Dependency (SCA) Report", style="bold #ea580c")
+        title = args.get("title", "")
+        if title:
+            text.append("\n\n")
+            text.append("Title: ", style=FIELD_STYLE)
+            text.append(title)
+
+        warning = result.get("warning")
+        if result.get("success") is False:
+            errors = result.get("errors")
+            detail = (
+                "; ".join(errors) if isinstance(errors, list) and errors else result.get("error")
+            )
+            label, style = "✗ Not created: ", "bold #dc2626"
+            fallback = "Report was not created."
+        else:
+            detail = warning
+            label, style = "⚠ Not persisted: ", "bold #d97706"
+            fallback = "Report could not be persisted."
+        text.append("\n\n")
+        text.append(label, style=style)
+        text.append(str(detail or fallback))
+
+        padded = Text()
+        padded.append("\n\n")
+        padded.append_text(text)
+        padded.append("\n\n")
+        return Static(padded, classes=cls.get_css_classes("failed"))
+
+    @classmethod
+    def render(cls, tool_data: dict[str, Any]) -> Static:  # noqa: PLR0912, PLR0915
+        args = tool_data.get("args", {})
+        result = tool_data.get("result", {})
+
+        if isinstance(result, dict) and (result.get("success") is False or result.get("warning")):
+            return cls._render_unsuccessful(args, result)
+
+        title = args.get("title", "")
+        description = args.get("description", "")
+        impact = args.get("impact", "")
+        target = args.get("target", "")
+        technical_analysis = args.get("technical_analysis", "")
+        remediation_steps = args.get("remediation_steps", "")
+        assumptions = args.get("assumptions", "")
+
+        package_name = args.get("package_name", "")
+        package_ecosystem = args.get("package_ecosystem", "")
+        installed_version = args.get("installed_version", "")
+        fixed_version = args.get("fixed_version", "")
+        cve = args.get("cve", "")
+        cwe = args.get("cwe", "")
+        advisory_cvss = args.get("advisory_cvss")
+        fix_effort = args.get("fix_effort", "")
+
+        severity = ""
+        if isinstance(result, dict):
+            severity = result.get("severity", "")
+
+        text = Text()
+        text.append("📦 ")
+        text.append("Dependency (SCA) Report", style="bold #ea580c")
+
+        if title:
+            text.append("\n\n")
+            text.append("Title: ", style=FIELD_STYLE)
+            text.append(title)
+
+        if severity:
+            text.append("\n\n")
+            text.append("Severity: ", style=FIELD_STYLE)
+            severity_color = cls.SEVERITY_COLORS.get(severity.lower(), "#6b7280")
+            text.append(severity.upper(), style=f"bold {severity_color}")
+
+        if advisory_cvss is not None:
+            text.append("\n\n")
+            text.append("Advisory CVSS: ", style=FIELD_STYLE)
+            try:
+                score = float(advisory_cvss)
+                text.append(str(score), style=f"bold {cls._get_cvss_color(score)}")
+            except (TypeError, ValueError):
+                text.append(str(advisory_cvss), style=DIM_STYLE)
+
+        if cve:
+            text.append("\n\n")
+            text.append("CVE: ", style=FIELD_STYLE)
+            text.append(cve)
+
+        if cwe:
+            text.append("\n\n")
+            text.append("CWE: ", style=FIELD_STYLE)
+            text.append(cwe)
+
+        if package_name:
+            text.append("\n\n")
+            text.append("Package: ", style=FIELD_STYLE)
+            text.append(package_name, style=FILE_STYLE)
+            if package_ecosystem:
+                text.append(f" ({package_ecosystem})", style=DIM_STYLE)
+
+        if installed_version:
+            text.append("\n\n")
+            text.append("Installed: ", style=FIELD_STYLE)
+            text.append(installed_version, style=BEFORE_STYLE)
+            if fixed_version:
+                text.append("  →  ", style=DIM_STYLE)
+                text.append("Fixed: ", style=FIELD_STYLE)
+                text.append(fixed_version, style=AFTER_STYLE)
+
+        if fix_effort:
+            text.append("\n\n")
+            text.append("Fix Effort: ", style=FIELD_STYLE)
+            text.append(fix_effort)
+
+        if target:
+            text.append("\n\n")
+            text.append("Target: ", style=FIELD_STYLE)
+            text.append(target)
+
+        for label, value in [
+            ("Description", description),
+            ("Impact", impact),
+            ("Technical Analysis", technical_analysis),
+            ("Assumptions", assumptions),
+            ("Remediation", remediation_steps),
+        ]:
+            if value:
+                text.append("\n\n")
+                text.append(label, style=FIELD_STYLE)
+                text.append("\n")
+                text.append(value)
+
+        if not title:
+            text.append("\n  ")
+            text.append("Creating dependency report...", style="dim")
+
+        padded = Text()
+        padded.append("\n\n")
+        padded.append_text(text)
+        padded.append("\n\n")
+
+        css_classes = cls.get_css_classes("completed")
+        return Static(padded, classes=css_classes)
@@ -4,6 +4,7 @@ from __future__ import annotations

 import json
 import logging
+import re
 from typing import TYPE_CHECKING, Any

 from agents.model_settings import ModelSettings
@@ -15,6 +16,7 @@ from strix.config.models import (
    DEFAULT_MODEL_RETRY,
    StrixProvider,
    configure_sdk_model_defaults,
+    request_timeout_extra_args,
 )
 from strix.report.state import get_global_report_state

@@ -51,6 +53,11 @@ CRITICAL DEDUPLICATION RULES:
   - One report is more thorough than another
   - Minor variations in technical analysis

+4. DEPENDENCY-CVE reports use package identity:
+   - Same CVE and same package/ecosystem is a duplicate
+   - Same CVE but different package/ecosystem is NOT a duplicate
+   - Same package/ecosystem but different CVE is NOT a duplicate
+
 COMPARISON GUIDELINES:
 - Focus on the technical root cause, not surface-level similarities
 - Same vulnerability type (SQLi, XSS) doesn't mean duplicate - location matters
@@ -101,6 +108,8 @@ def _prepare_report_for_comparison(report: dict[str, Any]) -> dict[str, Any]:
        "poc_description",
        "endpoint",
        "method",
+        "cve",
+        "dependency_metadata",
    ]

    cleaned = {}
@@ -114,6 +123,112 @@ def _prepare_report_for_comparison(report: dict[str, Any]) -> dict[str, Any]:
    return cleaned


+def _dependency_identity(report: dict[str, Any]) -> tuple[str, str, str] | None:
+    metadata = report.get("dependency_metadata")
+    if not isinstance(metadata, dict):
+        return None
+
+    raw_cve = report.get("cve")
+    raw_package = metadata.get("package_name")
+    if not raw_cve or not raw_package:
+        return None
+
+    cve = str(raw_cve).strip().upper()
+    ecosystem = str(metadata.get("package_ecosystem") or "").strip().lower()
+    package_name = str(raw_package).strip().lower()
+    if not cve or not package_name:
+        return None
+    return cve, ecosystem, package_name
+
+
+def _report_cve(report: dict[str, Any]) -> str:
+    return str(report.get("cve") or "").strip().upper()
+
+
+def _legacy_report_mentions_package(
+    report: dict[str, Any],
+    *,
+    ecosystem: str,
+    package_name: str,
+) -> bool:
+    fields = [
+        "title",
+        "description",
+        "impact",
+        "target",
+        "technical_analysis",
+        "poc_description",
+        "evidence",
+    ]
+    haystack = " ".join(str(report.get(field) or "") for field in fields).lower()
+    package_pattern = rf"(?<![\w@./-]){re.escape(package_name)}(?![\w@./-])"
+    if re.search(package_pattern, haystack) is None:
+        return False
+    if not ecosystem:
+        return True
+    ecosystem_pattern = rf"(?<![\w@./-]){re.escape(ecosystem)}(?![\w@./-])"
+    return re.search(ecosystem_pattern, haystack) is not None
+
+
+def _check_dependency_duplicate(
+    candidate: dict[str, Any],
+    existing_reports: list[dict[str, Any]],
+) -> dict[str, Any] | None:
+    candidate_identity = _dependency_identity(candidate)
+    if candidate_identity is None:
+        return None
+
+    cve, ecosystem, package_name = candidate_identity
+    found_legacy_same_cve = False
+    for report in existing_reports:
+        report_identity = _dependency_identity(report)
+        if report_identity is not None:
+            report_cve, report_ecosystem, report_package_name = report_identity
+            if (report_cve, report_package_name) != (cve, package_name):
+                continue
+            if report_ecosystem == ecosystem:
+                return {
+                    "is_duplicate": True,
+                    "duplicate_id": str(report.get("id") or "")[:64],
+                    "confidence": 1.0,
+                    "reason": "Same dependency CVE/package identity",
+                }
+            if not report_ecosystem or not ecosystem:
+                return {
+                    "is_duplicate": True,
+                    "duplicate_id": str(report.get("id") or "")[:64],
+                    "confidence": 1.0,
+                    "reason": "Same dependency CVE/package identity with missing ecosystem",
+                }
+            continue
+
+        if _report_cve(report) != cve:
+            continue
+        found_legacy_same_cve = True
+        if _legacy_report_mentions_package(
+            report,
+            ecosystem=ecosystem,
+            package_name=package_name,
+        ):
+            return {
+                "is_duplicate": True,
+                "duplicate_id": str(report.get("id") or "")[:64],
+                "confidence": 1.0,
+                "reason": "Same dependency CVE/package identity in legacy report",
+            }
+
+    if found_legacy_same_cve:
+        return None
+
+    package_label = f"{ecosystem}/{package_name}" if ecosystem else package_name
+    return {
+        "is_duplicate": False,
+        "duplicate_id": "",
+        "confidence": 1.0,
+        "reason": f"No existing dependency report for {cve} in {package_label}",
+    }
+
+
 def _parse_dedupe_response(content: str) -> dict[str, Any]:
    text = content.strip()
    if text.startswith("```"):
@@ -165,6 +280,10 @@ async def check_duplicate(
            "reason": "No existing reports to compare against",
        }

+    dependency_duplicate = _check_dependency_duplicate(candidate, existing_reports)
+    if dependency_duplicate is not None:
+        return dependency_duplicate
+
    try:
        settings = load_settings()
        model_name = settings.llm.model
@@ -192,7 +311,11 @@ async def check_duplicate(
        response = await model.get_response(
            system_instructions=DEDUPE_SYSTEM_PROMPT,
            input=user_msg,
-            model_settings=ModelSettings(retry=DEFAULT_MODEL_RETRY, include_usage=True),
+            model_settings=ModelSettings(
+                retry=DEFAULT_MODEL_RETRY,
+                include_usage=True,
+                extra_args=request_timeout_extra_args(settings.llm.timeout),
+            ),
            tools=[],
            output_schema=None,
            handoffs=[],
@@ -135,6 +135,10 @@ class ReportState:
        self._sarif_repo_ctx: dict[str, Any] | None = None
        self._sarif_repo_ctx_ready: bool = False

+        self.posthog_scan_ended_sent: bool = False
+        self.scarf_scan_ended_sent: bool = False
+        self.scan_ended_exit_reason: str | None = None
+
    def get_run_dir(self) -> Path:
        if self._run_dir is None:
            run_dir_name = self.run_name if self.run_name else self.run_id
@@ -212,6 +216,9 @@ class ReportState:
        poc_description: str | None = None,
        poc_script_code: str | None = None,
        remediation_steps: str | None = None,
+        evidence: str | None = None,
+        assumptions: str | None = None,
+        fix_effort: str | None = None,
        cvss: float | None = None,
        cvss_breakdown: dict[str, str] | None = None,
        endpoint: str | None = None,
@@ -219,6 +226,9 @@ class ReportState:
        cve: str | None = None,
        cwe: str | None = None,
        code_locations: list[dict[str, Any]] | None = None,
+        fix_pr_body: str | None = None,
+        finding_class: str | None = None,
+        dependency_metadata: dict[str, str] | None = None,
        agent_id: str | None = None,
        agent_name: str | None = None,
    ) -> str:
@@ -245,6 +255,12 @@ class ReportState:
            report["poc_script_code"] = poc_script_code.strip()
        if remediation_steps:
            report["remediation_steps"] = remediation_steps.strip()
+        if evidence:
+            report["evidence"] = evidence.strip()
+        if assumptions:
+            report["assumptions"] = assumptions.strip()
+        if fix_effort:
+            report["fix_effort"] = fix_effort.strip().lower()
        if cvss is not None:
            report["cvss"] = cvss
        if cvss_breakdown:
@@ -259,6 +275,11 @@ class ReportState:
            report["cwe"] = cwe.strip()
        if code_locations:
            report["code_locations"] = code_locations
+        if fix_pr_body:
+            report["fix_pr_body"] = fix_pr_body.strip()
+        report["finding_class"] = (finding_class or "dynamic").strip().lower()
+        if dependency_metadata:
+            report["dependency_metadata"] = dependency_metadata
        if agent_id:
            report["agent_id"] = agent_id
        if agent_name:
@@ -266,8 +287,8 @@ class ReportState:

        self.vulnerability_reports.append(report)
        logger.info(f"Added vulnerability report: {report_id} - {title}")
-        posthog.finding(severity)
-        scarf.finding(severity)
+        posthog.finding(severity, cwe=cwe, is_cve=bool(cve))
+        scarf.finding(severity, cwe=cwe, is_cve=bool(cve))

        if self.vulnerability_found_callback:
            self.vulnerability_found_callback(report)
@@ -513,16 +534,10 @@ def litellm_cost_callback(
                cost = value

    if cost is None:
-        usage: Any = getattr(completion_response, "usage", None)
-        if usage is None and isinstance(completion_response, dict):
-            usage = cast("dict[str, Any]", completion_response).get("usage")
-        usage_cost: Any
-        if isinstance(usage, dict):
-            usage_cost = cast("dict[str, Any]", usage).get("cost")
-        else:
-            usage_cost = getattr(usage, "cost", None)
-        if isinstance(usage_cost, int | float) and usage_cost > 0:
-            cost = float(usage_cost)
+        cost = _usage_reported_cost(completion_response)
+
+    if cost is None:
+        cost = _estimate_response_cost(kwargs, completion_response)

    if cost is None or cost <= 0:
        return
@@ -533,3 +548,101 @@ def litellm_cost_callback(
        report_state.record_observed_llm_cost(cost)
    except Exception:
        logger.exception("Failed to record observed LiteLLM cost")
+
+
+def _usage_reported_cost(completion_response: Any) -> float | None:
+    """Provider-reported cost from the ``usage`` block (e.g. OpenRouter).
+
+    Non-BYOK responses charge everything to ``usage.cost``. BYOK responses
+    charge only the OpenRouter fee to ``usage.cost`` (often 0) and report the
+    provider charge in ``usage.cost_details.upstream_inference_cost``, so the
+    true BYOK total is the sum of the two.
+    """
+    usage: Any = getattr(completion_response, "usage", None)
+    if usage is None and isinstance(completion_response, dict):
+        usage = cast("dict[str, Any]", completion_response).get("usage")
+    if usage is None:
+        return None
+
+    def _field(container: Any, name: str) -> Any:
+        if isinstance(container, dict):
+            return cast("dict[str, Any]", container).get(name)
+        return getattr(container, name, None)
+
+    total = 0.0
+    usage_cost = _field(usage, "cost")
+    if isinstance(usage_cost, int | float) and usage_cost > 0:
+        total += float(usage_cost)
+
+    if bool(_field(usage, "is_byok")):
+        upstream = _field(_field(usage, "cost_details"), "upstream_inference_cost")
+        if isinstance(upstream, int | float) and upstream > 0:
+            total += float(upstream)
+
+    return total if total > 0 else None
+
+
+def _estimate_response_cost(kwargs: Any, completion_response: Any) -> float | None:
+    """Best-effort LiteLLM cost-map estimate when no provider-reported cost exists.
+
+    LiteLLM strips provider cost fields when rebuilding streamed responses and
+    returns no ``response_cost`` for models missing from its cost map, so try
+    the provider-prefixed name, the raw name, and the bare model name.
+    """
+    from litellm import completion_cost
+
+    model = kwargs.get("model") if isinstance(kwargs, dict) else None
+    if not isinstance(model, str) or not model:
+        if isinstance(completion_response, dict):
+            model = cast("dict[str, Any]", completion_response).get("model")
+        else:
+            model = getattr(completion_response, "model", None)
+    if not isinstance(model, str) or not model:
+        return None
+
+    provider = None
+    litellm_params = kwargs.get("litellm_params") if isinstance(kwargs, dict) else None
+    if isinstance(litellm_params, dict):
+        provider = litellm_params.get("custom_llm_provider")
+
+    usage_payload = _usage_payload(completion_response)
+    if usage_payload is None:
+        return None
+
+    candidates: list[str] = []
+    if isinstance(provider, str) and provider and not model.startswith(f"{provider}/"):
+        candidates.append(f"{provider}/{model}")
+    candidates.append(model)
+    if "/" in model:
+        candidates.append(model.rsplit("/", 1)[-1])
+
+    for candidate in candidates:
+        try:
+            value = completion_cost(
+                completion_response={"model": candidate, "usage": usage_payload},
+                model=candidate,
+            )
+        except Exception:  # nosec B112  # noqa: BLE001, S112
+            continue
+        if isinstance(value, int | float) and value > 0:
+            return float(value)
+    return None
+
+
+def _usage_payload(completion_response: Any) -> dict[str, Any] | None:
+    """Token counts as a plain dict, detached from the response's provider metadata."""
+    usage: Any = getattr(completion_response, "usage", None)
+    if usage is None and isinstance(completion_response, dict):
+        usage = cast("dict[str, Any]", completion_response).get("usage")
+    if usage is None:
+        return None
+    if hasattr(usage, "model_dump"):
+        usage = usage.model_dump()
+    if not isinstance(usage, dict):
+        return None
+    payload = cast("dict[str, Any]", usage)
+    if not payload.get("total_tokens") and not (
+        payload.get("prompt_tokens") or payload.get("completion_tokens")
+    ):
+        return None
+    return payload
@@ -124,8 +124,13 @@ def render_vulnerability_md(report: dict[str, Any]) -> str:  # noqa: PLR0912, PL
        f"**Found:** {report.get('timestamp', 'unknown')}",
    ]

+    dep_meta = report.get("dependency_metadata") or {}
    metadata: list[tuple[str, Any]] = [
        ("Target", report.get("target")),
+        ("Package", dep_meta.get("package_name")),
+        ("Ecosystem", dep_meta.get("package_ecosystem")),
+        ("Installed Version", dep_meta.get("installed_version")),
+        ("Fixed Version", dep_meta.get("fixed_version")),
        ("Endpoint", report.get("endpoint")),
        ("Method", report.get("method")),
        ("CVE", report.get("cve")),
@@ -134,6 +139,8 @@ def render_vulnerability_md(report: dict[str, Any]) -> str:  # noqa: PLR0912, PL
    cvss = report.get("cvss")
    if cvss is not None:
        metadata.append(("CVSS", cvss))
+    if report.get("fix_effort"):
+        metadata.append(("Fix Effort", str(report["fix_effort"]).title()))
    for label, value in metadata:
        if value:
            lines.append(f"**{label}:** {value}")
@@ -143,6 +150,11 @@ def render_vulnerability_md(report: dict[str, Any]) -> str:  # noqa: PLR0912, PL
    lines.append(report.get("description") or "No description provided.")
    lines.append("")

+    if report.get("evidence"):
+        lines.append("## Evidence\n")
+        lines.append(str(report["evidence"]))
+        lines.append("")
+
    if report.get("impact"):
        lines.append("## Impact\n")
        lines.append(str(report["impact"]))
@@ -194,4 +206,9 @@ def render_vulnerability_md(report: dict[str, Any]) -> str:  # noqa: PLR0912, PL
        lines.append(str(report["remediation_steps"]))
        lines.append("")

+    if report.get("assumptions"):
+        lines.append("## Assumptions\n")
+        lines.append(str(report["assumptions"]))
+        lines.append("")
+
    return "\n".join(lines)
@@ -10,6 +10,7 @@ exposed-port URL for all subsequent SDK calls.
 from __future__ import annotations

 import asyncio
+import contextlib
 import json
 import logging
 from typing import TYPE_CHECKING
@@ -93,9 +94,16 @@ async def bootstrap_caido(
    client = Client(host_url, auth=TokenAuthOptions(token=access_token))
    await client.connect()

-    project = await client.project.create(
-        CreateProjectOptions(name="sandbox", temporary=True),
-    )
-    await client.project.select(project.id)
+    try:
+        project = await client.project.create(
+            CreateProjectOptions(name="sandbox", temporary=True),
+        )
+        await client.project.select(project.id)
+    except BaseException:
+        # The connected client never reaches the session bundle if project
+        # setup fails, so close it here to avoid leaking the transport.
+        with contextlib.suppress(Exception):
+            await client.aclose()
+        raise
    logger.info("Caido project selected: %s", project.id)
    return client
@@ -24,20 +24,25 @@ from __future__ import annotations

 import contextlib
 import logging
+import os
 import uuid
-from typing import Any
+from typing import Any, cast

+from agents.sandbox.errors import ExposedPortUnavailableError
 from agents.sandbox.manifest import Manifest
 from agents.sandbox.sandboxes.docker import (
    DockerSandboxClient,
+    DockerSandboxSession,
    _build_docker_volume_mounts,
    _docker_port_key,
    _manifest_requires_fuse,
    _manifest_requires_sys_admin,
 )
 from agents.sandbox.session.sandbox_session import SandboxSession
+from agents.sandbox.types import ExposedPortEndpoint
 from docker import errors as docker_errors  # type: ignore[import-untyped, unused-ignore]
 from docker.models.containers import Container  # type: ignore[import-untyped, unused-ignore]
+from docker.types import LogConfig  # type: ignore[import-untyped, unused-ignore]
 from docker.types import Mount as DockerSDKMount  # type: ignore[import-untyped, unused-ignore]
 from docker.utils import parse_repository_tag  # type: ignore[import-untyped, unused-ignore]
 from requests.exceptions import RequestException
@@ -46,10 +51,107 @@ from requests.exceptions import RequestException
 logger = logging.getLogger(__name__)


+_SANDBOX_NETWORK_ENV = "STRIX_DOCKER_SANDBOX_NETWORK"
+
+
+def _sandbox_network() -> str | None:
+    value = os.environ.get(_SANDBOX_NETWORK_ENV, "").strip()
+    return value or None
+
+
+def _apply_sandbox_network(create_kwargs: dict[str, Any]) -> None:
+    network = _sandbox_network()
+    if network:
+        create_kwargs["network"] = network
+        create_kwargs.pop("ports", None)
+
+
+def _apply_resource_limits(create_kwargs: dict[str, Any]) -> None:
+    """Apply optional cgroup resource caps from the environment. Unset/blank
+    values leave docker's default (unbounded), so this is opt-in per host."""
+    mem_limit = os.environ.get("STRIX_SANDBOX_MEM_LIMIT", "").strip()
+    if mem_limit:
+        create_kwargs["mem_limit"] = mem_limit
+
+    shm_size = os.environ.get("STRIX_SANDBOX_SHM_SIZE", "").strip()
+    if shm_size:
+        create_kwargs["shm_size"] = shm_size
+
+    cpus = os.environ.get("STRIX_SANDBOX_CPUS", "").strip()
+    if cpus:
+        with contextlib.suppress(ValueError, OverflowError):
+            nano_cpus = int(float(cpus) * 1_000_000_000)
+            if 0 < nano_cpus <= 2**63 - 1:
+                create_kwargs["nano_cpus"] = nano_cpus
+
+    pids_limit = os.environ.get("STRIX_SANDBOX_PIDS_LIMIT", "").strip()
+    if pids_limit:
+        with contextlib.suppress(ValueError):
+            create_kwargs["pids_limit"] = int(pids_limit)
+
+
+def _apply_log_limits(create_kwargs: dict[str, Any]) -> None:
+    """Bound the container's json-file log so a runaway process in the sandbox
+    (e.g. a tool that busy-loops writing to stdout) cannot fill the host disk
+    and take the Docker daemon down with it.
+
+    Unlike the cgroup caps above, this defaults **on** — docker's own default
+    is an unbounded json-file, which is unsafe for an autonomous agent that
+    executes arbitrary commands. ``max-file`` rotation means the on-disk cap is
+    ``max-size * max-file``. Set ``STRIX_SANDBOX_LOG_MAX_SIZE`` to ``0``/``off``
+    to opt back out to docker's default."""
+    max_size = os.environ.get("STRIX_SANDBOX_LOG_MAX_SIZE", "50m").strip()
+    if max_size.lower() in ("0", "off", "none", "unlimited"):
+        return
+    max_file = os.environ.get("STRIX_SANDBOX_LOG_MAX_FILE", "3").strip() or "3"
+    create_kwargs["log_config"] = LogConfig(
+        type=LogConfig.types.JSON,
+        config={"max-size": max_size, "max-file": max_file},
+    )
+
+
+class StrixDockerSandboxSession(DockerSandboxSession):
+    sandbox_network: str = ""
+
+    async def _resolve_exposed_port(self, port: int) -> ExposedPortEndpoint:
+        try:
+            self._container.reload()
+        except docker_errors.APIError as e:
+            raise ExposedPortUnavailableError(
+                port=port,
+                exposed_ports=self.state.exposed_ports,
+                reason="backend_unavailable",
+                context={
+                    "backend": "docker",
+                    "detail": "container_reload_failed",
+                    "network": self.sandbox_network,
+                },
+                cause=e,
+            ) from e
+
+        attrs = getattr(self._container, "attrs", {}) or {}
+        networks = attrs.get("NetworkSettings", {}).get("Networks", {})
+        endpoint = networks.get(self.sandbox_network) or {}
+        ip = endpoint.get("IPAddress") or endpoint.get("GlobalIPv6Address")
+        if not isinstance(ip, str) or not ip:
+            raise ExposedPortUnavailableError(
+                port=port,
+                exposed_ports=self.state.exposed_ports,
+                reason="backend_unavailable",
+                context={
+                    "backend": "docker",
+                    "detail": "container_not_on_network",
+                    "network": self.sandbox_network,
+                },
+            )
+        host = f"[{ip}]" if ":" in ip else ip
+        return ExposedPortEndpoint(host=host, port=port, tls=False)
+
+
 class StrixDockerSandboxClient(DockerSandboxClient):
    # Host directories to bind-mount into the container, set by the docker
    # backend before ``create()``. Each item is ``{source, target, read_only}``.
-    strix_bind_mounts: list[dict[str, Any]] = []  # overridden per-instance in backends.py
+    strix_bind_mounts: list[dict[str, Any]] | None = None

    async def _create_container(
        self,
@@ -117,6 +219,10 @@ class StrixDockerSandboxClient(DockerSandboxClient):
        extra_hosts = create_kwargs.setdefault("extra_hosts", {})
        extra_hosts["host.docker.internal"] = "host-gateway"

+        _apply_sandbox_network(create_kwargs)
+        _apply_resource_limits(create_kwargs)
+        _apply_log_limits(create_kwargs)
+
        # Strix injection: host bind mounts (e.g. large repos passed via --mount)
        # that bypass the SDK's file-by-file LocalDir copy.
        bind_mounts = getattr(self, "strix_bind_mounts", ())
@@ -146,6 +252,15 @@ class StrixDockerSandboxClient(DockerSandboxClient):
        )
        return container

+    async def create(self, **kwargs: Any) -> SandboxSession:
+        session = await super().create(**kwargs)
+        network = _sandbox_network()
+        inner = session._inner
+        if network and isinstance(inner, DockerSandboxSession):
+            inner.__class__ = StrixDockerSandboxSession
+            cast("StrixDockerSandboxSession", inner).sandbox_network = network
+        return session
+
    async def delete(self, session: SandboxSession) -> SandboxSession:
        container_id = getattr(getattr(session._inner, "state", None), "container_id", None)
        if container_id:
@@ -0,0 +1,120 @@
+"""Symlink-safe staging for ``LocalDir`` manifest uploads.
+
+The sandbox SDK's ``LocalDir`` walker refuses to copy symlinks at all — it
+raises ``LocalDirReadError(reason="symlink_not_supported")`` on the first one
+as a path-escape / TOCTOU safeguard. Real source trees (especially JS/TS
+monorepos with workspace or shared-config links) routinely commit symlinks, so
+handing such a tree straight to ``LocalDir`` aborts the upload before the agent
+even starts.
+
+:func:`stage_symlink_safe_dir` returns a path that is always safe to hand to
+``LocalDir``:
+
+* a tree with no symlinks is used as-is (no copy);
+* otherwise the tree is copied into a temp directory with symlinks resolved:
+
+  - a link whose target stays inside the tree is *dereferenced* (its target
+    content is materialized in place), so the agent still sees the file;
+  - a link that escapes the tree, dangles, or forms a cycle is *dropped* and
+    never followed. Refusing to follow out-of-tree links preserves the walker's
+    path-escape safety and keeps host/out-of-tree content from leaking into the
+    (hostile) sandbox.
+
+Regular files are hard-linked when possible (falling back to a copy across
+devices), so the staged tree adds negligible disk for the non-symlink bulk.
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import shutil
+import tempfile
+from pathlib import Path
+
+
+logger = logging.getLogger(__name__)
+
+_STAGING_PREFIX = "strix-localdir-"
+
+
+def _is_within(target: Path, root: Path) -> bool:
+    """Return whether ``target`` is ``root`` itself or nested under it."""
+    if target == root:
+        return True
+    try:
+        target.relative_to(root)
+    except ValueError:
+        return False
+    return True
+
+
+def tree_has_symlink(root: Path) -> bool:
+    """Return whether ``root`` contains any symlink (file or directory)."""
+    for dirpath, dirnames, filenames in os.walk(root, followlinks=False):
+        base = Path(dirpath)
+        for name in (*dirnames, *filenames):
+            if (base / name).is_symlink():
+                return True
+    return False
+
+
+def _link_or_copy(src: Path, dst: Path) -> None:
+    """Hard-link ``src`` to ``dst``, falling back to a content copy."""
+    try:
+        os.link(src, dst)
+    except OSError:
+        shutil.copy2(src, dst, follow_symlinks=True)
+
+
+def _stage_dir(src: Path, dst: Path, root: Path, seen: frozenset[Path]) -> None:
+    dst.mkdir(parents=True, exist_ok=True)
+    for entry in os.scandir(src):
+        entry_path = Path(entry.path)
+        dest_path = dst / entry.name
+
+        if entry.is_symlink():
+            target = Path(os.path.realpath(entry_path))
+            if not _is_within(target, root):
+                logger.warning("staging: dropping out-of-tree symlink %s -> %s", entry_path, target)
+                continue
+            if not target.exists():
+                logger.warning("staging: dropping dangling symlink %s", entry_path)
+                continue
+            if target in seen:
+                logger.warning("staging: dropping cyclic symlink %s -> %s", entry_path, target)
+                continue
+            if target.is_dir():
+                _stage_dir(target, dest_path, root, seen | {target})
+            else:
+                _link_or_copy(target, dest_path)
+        elif entry.is_dir(follow_symlinks=False):
+            _stage_dir(entry_path, dest_path, root, seen)
+        elif entry.is_file(follow_symlinks=False):
+            _link_or_copy(entry_path, dest_path)
+        else:
+            # Sockets, FIFOs, devices — not part of a source tree; skip.
+            logger.debug("staging: skipping non-regular entry %s", entry_path)
+
+
+def stage_symlink_safe_dir(src_root: Path) -> tuple[Path, Path | None]:
+    """Return ``(upload_path, staged_temp)`` for uploading ``src_root``.
+
+    ``upload_path`` is safe to hand to ``LocalDir``. When the tree contains no
+    symlinks it is ``src_root`` itself and ``staged_temp`` is ``None``.
+    Otherwise a symlink-safe copy is materialized in a temp directory and both
+    returned values point at it; the caller owns removing ``staged_temp`` once
+    the upload completes.
+    """
+    root = src_root.resolve()
+    if not tree_has_symlink(root):
+        return root, None
+
+    staged = Path(tempfile.mkdtemp(prefix=_STAGING_PREFIX))
+    try:
+        _stage_dir(root, staged, root, frozenset({root}))
+    except OSError:
+        shutil.rmtree(staged, ignore_errors=True)
+        raise
+    logger.info("staging: materialized symlink-safe copy of %s at %s", root, staged)
+    return staged, staged
@@ -3,6 +3,7 @@
 from __future__ import annotations

 import logging
+import shutil
 from pathlib import Path
 from typing import Any

@@ -12,6 +13,7 @@ from agents.sandbox.manifest import Environment, Manifest
 from strix.config import load_settings
 from strix.runtime.backends import get_backend
 from strix.runtime.caido_bootstrap import bootstrap_caido
+from strix.runtime.local_dir_staging import stage_symlink_safe_dir


 logger = logging.getLogger(__name__)
@@ -29,16 +31,20 @@ _WORKSPACE_ROOT = "/workspace"

 def build_session_entries(
    local_sources: list[dict[str, Any]],
-) -> tuple[dict[str | Path, BaseEntry], list[dict[str, Any]]]:
+) -> tuple[dict[str | Path, BaseEntry], list[dict[str, Any]], list[Path]]:
    """Split local sources into copied manifest entries and host bind mounts.

    Sources flagged ``mount`` are bind-mounted read-only at
    ``/workspace/<workspace_subdir>`` (not added to the manifest, so the SDK
    does not stream them in file-by-file). Every other source becomes a
-    ``LocalDir`` entry copied into the container as before.
+    ``LocalDir`` entry copied into the container as before. Trees containing
+    symlinks (which the SDK's ``LocalDir`` walker refuses outright) are first
+    staged into a symlink-safe temp copy; those temp dirs are returned so the
+    caller can remove them once the upload completes.
    """
    entries: dict[str | Path, BaseEntry] = {}
    bind_mounts: list[dict[str, Any]] = []
+    staged_dirs: list[Path] = []
    for src in local_sources:
        ws_subdir = src.get("workspace_subdir") or ""
        host_path = src.get("source_path") or ""
@@ -54,8 +60,11 @@ def build_session_entries(
                }
            )
        else:
-            entries[ws_subdir] = LocalDir(src=resolved)
-    return entries, bind_mounts
+            upload_path, staged = stage_symlink_safe_dir(resolved)
+            if staged is not None:
+                staged_dirs.append(staged)
+            entries[ws_subdir] = LocalDir(src=upload_path)
+    return entries, bind_mounts, staged_dirs


 async def create_or_reuse(
@@ -75,7 +84,7 @@ async def create_or_reuse(
        logger.info("Reusing existing sandbox session for scan %s", scan_id)
        return cached

-    entries, bind_mounts = build_session_entries(local_sources)
+    entries, bind_mounts, staged_dirs = build_session_entries(local_sources)

    # Caido runs as an in-container sidecar; HTTP(S) traffic from any
    # process started via ``session.exec`` (the SDK's Shell tool, etc.)
@@ -106,12 +115,16 @@ async def create_or_reuse(
        backend_name,
        image,
    )
-    client, session = await backend(
-        image=image,
-        manifest=manifest,
-        exposed_ports=(_CONTAINER_CAIDO_PORT,),
-        bind_mounts=bind_mounts,
-    )
+    try:
+        client, session = await backend(
+            image=image,
+            manifest=manifest,
+            exposed_ports=(_CONTAINER_CAIDO_PORT,),
+            bind_mounts=bind_mounts,
+        )
+    finally:
+        for staged in staged_dirs:
+            shutil.rmtree(staged, ignore_errors=True)

    caido_endpoint = await session.resolve_exposed_port(_CONTAINER_CAIDO_PORT)
    scheme = "https" if caido_endpoint.tls else "http"
@@ -154,11 +167,19 @@ async def cleanup(scan_id: str) -> None:
        except Exception:  # noqa: BLE001
            logger.debug("cleanup(%s): caido_client.aclose() raised", scan_id, exc_info=True)

+    client = bundle["client"]
    try:
-        await bundle["client"].delete(bundle["session"])
+        await client.delete(bundle["session"])
        logger.info("Cleaned up sandbox session for scan %s", scan_id)
    except Exception:
        logger.exception(
            "cleanup(%s): client.delete raised; container may need manual reaping",
            scan_id,
        )
+
+    docker_client = getattr(client, "docker_client", None)
+    if docker_client is not None:
+        try:
+            docker_client.close()
+        except Exception:  # noqa: BLE001
+            logger.debug("cleanup(%s): docker_client.close() raised", scan_id, exc_info=True)
@@ -41,6 +41,7 @@ The skills are dynamically injected into the agent's system prompt, allowing it
 Notable source-aware skills:
 - `source_aware_whitebox` (coordination): white-box orchestration playbook
 - `source_aware_sast` (custom): semgrep/AST/secrets/supply-chain static triage workflow
+- `dependency_cve_scanning` (custom): trivy-based SCA workflow for reporting known dependency CVEs via `create_dependency_report`

 ---

@@ -1,7 +1,11 @@
 import logging
 import re
+import threading
+from collections import Counter
 from collections.abc import Iterator
+from pathlib import Path

+from strix.telemetry import posthog, scarf
 from strix.utils.resource_paths import get_strix_resource_path


@@ -10,20 +14,82 @@ logger = logging.getLogger(__name__)
 _FRONTMATTER_PATTERN = re.compile(r"^---\s*\n.*?\n---\s*\n", re.DOTALL)

 _INTERNAL_SKILL_CATEGORIES: frozenset[str] = frozenset({"scan_modes", "coordination"})
+_ROOT_SKILL_CATEGORY = "root"
+
+_EXTRA_SKILL_DIRS: list[Path] = []
+
+
+def register_skill_dir(path: str | Path) -> None:
+    """Add a directory searched for skills ahead of the built-in set.
+
+    The directory uses the same layout as the packaged skills
+    (``<root>/<category>/<name>.md``). Skills found in a registered
+    directory shadow packaged skills with the same relative path, so
+    callers can both add new skills and override existing ones without
+    editing the package. The most recently registered directory has the
+    highest precedence.
+    """
+    resolved = Path(path)
+    if resolved not in _EXTRA_SKILL_DIRS:
+        _EXTRA_SKILL_DIRS.append(resolved)
+        logger.info("Registered extra skill dir: %s", resolved)
+
+
+def registered_skill_dirs() -> tuple[Path, ...]:
+    """Return registered extra skill directories, highest precedence first."""
+    return tuple(reversed(_EXTRA_SKILL_DIRS))
+
+
+def skill_search_dirs() -> tuple[Path, ...]:
+    """All existing skill roots, highest precedence first (built-in last)."""
+    roots = [d for d in registered_skill_dirs() if d.is_dir()]
+    builtin = get_strix_resource_path("skills")
+    if builtin.is_dir():
+        roots.append(builtin)
+    return tuple(roots)


 def _iter_user_skill_files() -> Iterator[tuple[str, str]]:
    """Yield ``(category_name, skill_name)`` for every user-selectable skill."""
-    skills_dir = get_strix_resource_path("skills")
-    if not skills_dir.exists():
-        return
-    for category_dir in sorted(skills_dir.iterdir()):
-        if not category_dir.is_dir() or category_dir.name.startswith("__"):
-            continue
-        if category_dir.name in _INTERNAL_SKILL_CATEGORIES:
-            continue
-        for file_path in sorted(category_dir.glob("*.md")):
-            yield category_dir.name, file_path.stem
+    seen: set[tuple[str, str]] = set()
+    for skills_dir in skill_search_dirs():
+        for file_path in sorted(skills_dir.glob("*.md")):
+            if file_path.name.startswith("__") or file_path.name == "README.md":
+                continue
+            key = (_ROOT_SKILL_CATEGORY, file_path.stem)
+            if key in seen:
+                continue
+            seen.add(key)
+            yield key
+
+        for category_dir in sorted(skills_dir.iterdir()):
+            if not category_dir.is_dir() or category_dir.name.startswith("__"):
+                continue
+            if category_dir.name in _INTERNAL_SKILL_CATEGORIES:
+                continue
+            for file_path in sorted(category_dir.glob("*.md")):
+                key = (category_dir.name, file_path.stem)
+                if key in seen:
+                    continue
+                seen.add(key)
+                yield key
+
+
+def _is_selectable_root_skill_file(file_path: Path) -> bool:
+    return file_path.suffix == ".md" and not (
+        file_path.name.startswith("__") or file_path.name == "README.md"
+    )
+
+
+def _qualified_skill_file(skills_dir: Path, category: str, name: str) -> Path | None:
+    if category == _ROOT_SKILL_CATEGORY:
+        candidate = skills_dir / f"{name}.md"
+        if candidate.exists() and _is_selectable_root_skill_file(candidate):
+            return candidate
+        return None
+
+    candidate = skills_dir / category / f"{name}.md"
+    return candidate if candidate.exists() else None


 def get_all_skill_names() -> set[str]:
@@ -31,6 +97,54 @@ def get_all_skill_names() -> set[str]:
    return {name for _, name in _iter_user_skill_files()}


+def _get_all_skill_keys() -> set[str]:
+    keys: set[str] = set()
+    for category, name in _iter_user_skill_files():
+        keys.add(f"{category}/{name}")
+    return keys
+
+
+def _get_ambiguous_skill_names() -> set[str]:
+    counts = Counter(name for _, name in _iter_user_skill_files())
+    return {name for name, count in counts.items() if count > 1}
+
+
+def _qualified_skill_files(skill_name: str) -> list[Path]:
+    category, _, name = skill_name.partition("/")
+    for skills_dir in skill_search_dirs():
+        candidate = _qualified_skill_file(skills_dir, category, name)
+        if candidate is not None:
+            return [candidate]
+    return []
+
+
+def _bare_skill_files(skill_name: str) -> list[Path]:
+    seen: set[tuple[str, str]] = set()
+    candidates: list[Path] = []
+    for skills_dir in skill_search_dirs():
+        for category_dir in sorted(skills_dir.iterdir()):
+            if not category_dir.is_dir() or category_dir.name.startswith("__"):
+                continue
+            if category_dir.name in _INTERNAL_SKILL_CATEGORIES:
+                continue
+            key = (category_dir.name, skill_name)
+            if key in seen:
+                continue
+            candidate = category_dir / f"{skill_name}.md"
+            if candidate.exists():
+                seen.add(key)
+                candidates.append(candidate)
+
+        key = (_ROOT_SKILL_CATEGORY, skill_name)
+        if key in seen:
+            continue
+        root_candidate = _qualified_skill_file(skills_dir, _ROOT_SKILL_CATEGORY, skill_name)
+        if root_candidate is not None:
+            seen.add(key)
+            candidates.append(root_candidate)
+    return candidates
+
+
 def get_available_skills() -> dict[str, list[str]]:
    grouped: dict[str, list[str]] = {}
    for category, name in _iter_user_skill_files():
@@ -52,48 +166,63 @@ def validate_requested_skills(skill_list: list[str], max_skills: int = 5) -> str
    if not skill_list:
        return None
    available = get_all_skill_names()
-    invalid = sorted({s for s in skill_list if s not in available})
+    available_keys = _get_all_skill_keys()
+    invalid = sorted({s for s in skill_list if s not in available and s not in available_keys})
    if invalid:
        return f"Invalid skill name(s): {invalid}. Available skills: {sorted(available)}"
+    ambiguous = sorted({s for s in skill_list if "/" not in s} & _get_ambiguous_skill_names())
+    if ambiguous:
+        return (
+            f"Ambiguous skill name(s): {ambiguous}. Use category-qualified names from: "
+            f"{sorted(available_keys)}"
+        )
    return None


+def _track_skill_loaded(skill_name: str, file_path: Path) -> None:
+    builtin = get_strix_resource_path("skills")
+    if not file_path.is_relative_to(builtin):
+        skill_name = "custom"
+
+    def _send() -> None:
+        posthog.skill_loaded(skill_name)
+        scarf.skill_loaded(skill_name)
+
+    threading.Thread(target=_send, daemon=True).start()
+
+
+def _candidate_skill_files(skill_name: str) -> list[Path]:
+    """Resolve *skill_name* to effective matching files."""
+    if "/" in skill_name:
+        return _qualified_skill_files(skill_name)
+    return _bare_skill_files(skill_name)
+
+
 def load_skills(skill_names: list[str]) -> dict[str, str]:
    """Load skill markdown bodies (frontmatter stripped) by name.

-    Skill files live at ``strix/skills/<category>/<name>.md``. Names
-    can be ``"name"`` (any category), ``"category/name"``, or a bare
-    file at the skills root. Missing skills are logged and skipped.
+    Skill files live at ``strix/skills/<category>/<name>.md`` (or any
+    directory added via :func:`register_skill_dir`, searched first).
+    Names can be ``"name"`` (any category), ``"category/name"``, or a
+    bare file at the skills root. Missing skills are logged and skipped.
    """
-    skills_dir = get_strix_resource_path("skills")
-    if not skills_dir.exists():
+    search_dirs = skill_search_dirs()
+    if not search_dirs:
        return {}

-    by_category: dict[str, str] = {}
-    for category_dir in skills_dir.iterdir():
-        if not category_dir.is_dir() or category_dir.name.startswith("__"):
-            continue
-        for file_path in category_dir.glob("*.md"):
-            by_category[file_path.stem] = f"{category_dir.name}/{file_path.stem}.md"
-
    skill_content: dict[str, str] = {}
    for skill_name in skill_names:
-        rel_path: str | None
-        if "/" in skill_name:
-            rel_path = f"{skill_name}.md"
-        elif skill_name in by_category:
-            rel_path = by_category[skill_name]
-        elif (skills_dir / f"{skill_name}.md").exists():
-            rel_path = f"{skill_name}.md"
-        else:
-            rel_path = None
-
-        if rel_path is None or not (skills_dir / rel_path).exists():
+        candidates = _candidate_skill_files(skill_name)
+        if not candidates:
            logger.warning("Skill not found: %s", skill_name)
            continue
+        if len(candidates) > 1:
+            logger.warning("Ambiguous skill name %s; use a category-qualified name", skill_name)
+            continue
+        file_path = candidates[0]

        try:
-            content = (skills_dir / rel_path).read_text(encoding="utf-8")
+            content = file_path.read_text(encoding="utf-8")
        except (OSError, ValueError) as e:
            logger.warning("Failed to load skill %s: %s", skill_name, e)
            continue
@@ -101,6 +230,7 @@ def load_skills(skill_names: list[str]) -> dict[str, str]:
        var_name = skill_name.split("/")[-1]
        skill_content[var_name] = _FRONTMATTER_PATTERN.sub("", content).lstrip()
        logger.debug("Loaded skill: %s -> %s", skill_name, var_name)
+        _track_skill_loaded(var_name, file_path)

    logger.debug("load_skills: %d skill(s) resolved", len(skill_content))
    return skill_content
@@ -0,0 +1,138 @@
+---
+name: dependency-cve-scanning
+description: Supply-chain / SCA playbook — scan repository lockfiles for known dependency CVEs and report them with create_dependency_report (no dynamic PoC required)
+---
+
+# Dependency / Supply-Chain CVE Scanning (SCA)
+
+Use this skill on white-box / repository scans to make sure a repository pinning a
+**known-vulnerable dependency** is actually reported as a finding, instead of being
+discovered and then silently dropped because it cannot be dynamically exploited.
+
+Known-CVE dependency findings are a first-class deliverable. Report each one with
+the dedicated `create_dependency_report` tool.
+
+## Why this skill exists
+
+A vulnerable dependency pinned in a lockfile (e.g. `lodash@4.17.4` with a known
+prototype-pollution CVE) usually cannot be dynamically PoC'd from the outside —
+the vulnerable code path may not even be reachable from a running endpoint. The
+normal "no report without a dynamic PoC" rule would suppress it. For these
+findings the proof is the **lockfile entry + scanner output + published
+advisory**, not an exploit script. This is the one explicit exception to the
+dynamic-validation rule, and it exists only for `create_dependency_report`.
+
+## Scan procedure
+
+Run from the repo root and store output in the shared artifact directory used by
+the source-aware pass:
+
+```bash
+ART=/workspace/.strix-source-aware
+mkdir -p "$ART"
+
+# Record the vuln DB age so a stale DB is a visible signal, not a silent clean scan.
+trivy version --format json 2>/dev/null | tee "$ART/trivy-version.json"
+# inspect .VulnerabilityDB.UpdatedAt / NextUpdate
+
+# Lockfile/manifest -> known-CVE matching. Try a best-effort DB refresh first so a
+# sandbox with egress gets the freshest CVEs; if the update fails, fall back to the
+# cached DB instead of failing the scan. --offline-scan keeps per-package advisory
+# lookups offline.
+trivy fs --scanners vuln --timeout 30m --offline-scan \
+  --format json --output "$ART/trivy-sca.json" . \
+  || trivy fs --scanners vuln --timeout 30m --offline-scan --skip-db-update \
+       --format json --output "$ART/trivy-sca.json" . \
+  || true
+```
+
+If `.VulnerabilityDB.UpdatedAt` is more than a few weeks old (the sandbox had no
+egress to refresh it), treat it as a scan limitation and note it in the
+`assumptions` of dependency findings — a stale DB that still returns *some* results
+will not trip the "zero results is suspicious" heuristic, so its age is the only
+staleness signal.
+
+Trivy reads the lockfiles/manifests it finds, including:
+`package-lock.json`, `yarn.lock`, `pnpm-lock.yaml`, `poetry.lock`,
+`requirements.txt`, `Pipfile.lock`, `go.mod`/`go.sum`, `Gemfile.lock`,
+`pom.xml`/`gradle.lockfile`, `Cargo.lock`, `composer.lock`, etc.
+
+If trivy returns zero vulnerabilities on a repo with dependencies, treat it as
+suspicious: confirm the vuln DB is present (`trivy-version.json`) and that
+lockfiles exist.
+
+## Interpreting results
+
+For each entry under `.Results[].Vulnerabilities[]` in `trivy-sca.json`, collect:
+
+- `VulnerabilityID` — the CVE (or GHSA; prefer the CVE if both are present)
+- `PkgName` and `InstalledVersion` — the affected package + pinned version
+- `FixedVersion` — the version that resolves it
+- `Target` — the lockfile path it came from
+- `.Results[].Type` (e.g. `npm`, `pip`, `gomod`, `pom`, `gemspec`, `cargo`) — the
+  package ecosystem; normalize to the registry name lowercased (`npm`, `pypi`,
+  `go`, `maven`, `rubygems`, `cargo`, `composer`, `nuget`, ...)
+- `CVSS` — the published advisory base score
+- `PrimaryURL` / references — to verify the advisory
+
+Deduplicate by `(CVE, PkgName, InstalledVersion)`. File one
+`create_dependency_report` per CVE — do not batch multiple CVEs into one report.
+
+### Reachability is a confidence modifier, not a gate
+
+Do NOT suppress or downgrade a known CVE just because you could not prove the
+vulnerable code path is reachable. Report it, set `advisory_cvss` from the
+advisory, and use `assumptions` to note reachability (e.g. "the vulnerable
+`template()` API does not appear to be imported in application code, so practical
+exploitability is uncertain"). If you *can* show reachability or chain it into a
+dynamic exploit, do that and report it as a normal dynamic finding with
+`create_vulnerability_report` instead.
+
+## Reporting
+
+Report each confirmed known CVE with the dedicated `create_dependency_report`
+tool (NOT `create_vulnerability_report` — that tool is for dynamically validated
+findings and rejects empty PoC fields):
+
+- Set `cve` to the verified `CVE-YYYY-NNNNN` id (required). If you only have a
+  GHSA, look up the mapped CVE; if there is genuinely no CVE, do not report it
+  with this tool.
+- There are no PoC fields — `create_dependency_report` does not take
+  `poc_description` / `poc_script_code` / `code_locations`. The proof lives in
+  `description` and `technical_analysis` (scanner output + advisory).
+- **Always fill the structured dependency fields** (they power the dedicated
+  dependency-report card; do not leave them only in free-text):
+  - `package_name` — `PkgName` (required).
+  - `installed_version` — `InstalledVersion` (required).
+  - `package_ecosystem` — normalized ecosystem from `.Results[].Type` (lowercased,
+    e.g. `npm`, `pypi`, `go`, `maven`, `rubygems`, `cargo`) (required).
+  - `fixed_version` — `FixedVersion` (leave empty only if no fix is published).
+- Reference the repo-relative `Target` lockfile path in `description` /
+  `technical_analysis` (no leading slash) so the finding is traceable.
+- Put the concrete proof in `description` / `technical_analysis`: package name,
+  installed/affected version, fixed version, lockfile path, and the relevant
+  trivy output excerpt.
+- **Always set `advisory_cvss` to the published advisory base score (0.0–10.0).**
+  Severity is derived *solely* from this number: read it off the advisory (`CVSS`
+  in trivy output, or the NVD/GHSA page) and pass the real value. The tool rejects
+  a call that omits it, because guessing a score both inflates low CVEs and
+  deflates critical ones.
+- Set `cwe` to the most specific `CWE-NNN` when the advisory names one.
+- Do NOT cap severity at LOW just because there is no dynamic reproduction — use
+  the advisory score.
+- Use `assumptions` for reachability/exploitability caveats.
+
+Verify the CVE with `web_search` when available before reporting. Never guess or
+hallucinate a CVE id.
+
+## Anti-patterns
+
+- Do not report a dependency CVE with `create_vulnerability_report`; use
+  `create_dependency_report`.
+- Do not report a finding without a verified CVE id.
+- Do not batch multiple CVEs into one report.
+- Do not omit `advisory_cvss` — the tool rejects it, and it is the single input
+  that determines dependency severity.
+- Do not silently drop a known CVE because it lacks a dynamic PoC — that is the
+  exact failure this skill prevents.
+- Do not downgrade advisory severity for lack of dynamic reproduction.
@@ -121,6 +121,11 @@ trivy fs --scanners vuln,misconfig --timeout 30m --offline-scan \
  --format json --output /workspace/.strix-source-aware/trivy-fs.json . || true
 ```

+Known-CVE dependency findings are the one exception to the "report only after
+dynamic validation" rule below: report each one with `create_dependency_report`
+(not `create_vulnerability_report`), setting `advisory_cvss` from the published
+advisory. `load_skill(["dependency_cve_scanning"])` for the full SCA workflow.
+
 ## JavaScript-Side Coverage

 For frontends and Node services, layer these on top of the language-agnostic
@@ -0,0 +1,151 @@
+---
+name: asset-discovery
+description: Passive asset and attack-surface discovery via certificate transparency, TLS SAN pivoting, passive DNS, and ASN/IP enumeration to find hosts beyond subdomain brute force
+---
+
+# Asset Discovery
+
+Most engagements start from a small seed (one domain, one org name) but the real attack surface is far larger: forgotten hosts, staging/internal-named services, acquisitions, and infrastructure that never appears in a wordlist. Build a broad, deduplicated inventory using passive intelligence — certificate transparency, TLS certificate metadata, passive DNS, and ASN/IP data — then collapse it into a probed, classified attack surface. The aim is coverage and pivoting: every certificate, DNS record, and IP is a lead to more assets.
+
+Only use this skill when all subdomains and related assets of the target are in scope — broad discovery pulls in hosts far beyond the seed.
+
+## Attack Surface
+
+- Hosts discoverable via issued certificates (CT logs) but absent from DNS brute force
+- Internal/staging/pre-prod hostnames leaked in certificate SAN lists
+- Sibling and acquisition domains sharing certificates, ASNs, or IP ranges with the seed
+- Wildcard and short-lived certs revealing naming conventions (`*.internal.example.com`, `k8s-*`, `argocd.*`)
+- ASN-owned IP ranges hosting services with no DNS name at all
+- Virtual hosts co-located on shared IPs (multiple apps behind one address)
+- Non-HTTP services on discovered hosts (databases, brokers, admin ports)
+
+## High-Value Sources
+
+### Certificate Transparency (CT)
+
+CT logs record nearly every publicly-trusted certificate. Query by domain (matches SAN/CN) and by organization name.
+
+- **crt.sh** (free, no key):
+  - By domain incl. subdomains: `curl -s 'https://crt.sh/?q=%25.example.com&output=json' | jq -r '.[].name_value' | sed 's/^\*\.//' | sort -u`
+  - By organization: `https://crt.sh/?O=Example+Inc&output=json`
+- **Censys / Shodan / Fofa** (API keys): search certs by `parsed.names`, `parsed.subject.organization`, or a specific `fingerprint_sha256`, then pivot to every host serving that cert.
+- Cross-check multiple indexes (`certspotter`, Google CT, `chaos`) — no single log is complete.
+- **Wildcards** (`*.corp.example.com`) reveal internal naming schemes even when individual hosts resolve privately; use them to seed targeted guesses (`grafana.corp`, `ci.corp`, `vault.corp`).
+
+### TLS Certificate SAN/CN
+
+- **SAN expansion**: one cert often lists many hostnames (marketing + api + admin + internal) — extract every SAN, not just the queried name.
+- **Shared-cert pivot**: the same cert fingerprint served on multiple IPs ties disparate assets to one owner.
+- **Issuer/org pivot**: certs sharing `subject.organization`/`organizationalUnit` frequently belong to the same target.
+- **Active read** catches names never submitted to public CT: `echo | openssl s_client -connect HOST:443 -servername HOST 2>/dev/null | openssl x509 -noout -text | grep -A1 'Subject Alternative Name'`
+- **Internal leak signal**: SANs like `localhost`, `*.internal`, `*.svc.cluster.local`, `*.local`, or RFC1918-style names on a public cert expose internal naming and sometimes internal services fronted publicly.
+
+### Passive DNS
+
+- Forward-resolve every name (A/AAAA/CNAME); keep CNAME chains — they reveal third-party providers and CDNs.
+- **Reverse DNS (PTR)** on discovered IPs surfaces co-located hostnames.
+- **Historical/passive DNS** (SecurityTrails, VirusTotal, `chaos`, passivedns providers) recovers names that no longer resolve but may still front live infra.
+
+### ASN & IP Ranges
+
+- Map a known IP to its ASN and netblock: `whois -h whois.cymru.com " -v <IP>"` or a BGP/ASN lookup.
+- If the org runs its own ASN, enumerate all announced prefixes and treat them as candidate assets.
+- For cloud-hosted targets the IP belongs to the provider, not the org — pivot via cert/vhost instead of netblock.
+
+## Recommended Tooling
+
+Prefer the projectdiscovery suite (already available in the sandbox and pipeline-friendly with JSON output):
+
+- **`subfinder`** — passive subdomain aggregation across many sources incl. CT: `subfinder -d example.com -all -recursive -silent -oJ -o subs.jsonl`
+- **`tlsx`** — TLS/cert data at scale; grab SANs and issuer/org to pivot: `tlsx -l hosts.txt -san -cn -tls-version -json -o tls.jsonl`
+- **`uncover`** — query Shodan/Censys/Fofa/Quake/crt.sh engines from one CLI: `uncover -q 'ssl:"Example Inc"' -e shodan,censys,fofa -json`
+- **`asnmap`** — org/domain/ASN → CIDR ranges: `asnmap -d example.com -json` / `asnmap -org "Example Inc"`
+- **`mapcidr`** — expand/aggregate CIDRs into host lists for probing: `mapcidr -cidr 192.0.2.0/24 -o hosts.txt`
+- **`dnsx`** — fast resolution, PTR, and wildcard filtering: `dnsx -l names.txt -a -aaaa -cname -ptr -resp -json -o dns.jsonl`
+- **`httpx`** — live probing + cert grab in one pass (see methodology).
+- **`naabu`** — port sweep for non-HTTP services: `naabu -list hosts.txt -top-ports 100 -verify -silent`
+
+Also useful: **`amass`** (`amass intel`/`enum` for ASN, cert, and passive sources), **`cero`** (bulk SAN extraction from IPs/ranges), and direct **crt.sh** JSON queries when no keys are configured. Cross-source results — CT + passive DNS + `subfinder` together beat any single source.
+
+## Key Techniques
+
+### Iterative Seed Expansion
+
+Every new name, PTR result, CNAME target, and cert SAN becomes a fresh seed. Loop CT → SAN extraction → passive DNS → ASN/range expansion until the asset set stops growing.
+
+### Cert-Fingerprint Pivoting
+
+Search Censys/Shodan (or `uncover`) by a cert's `fingerprint_sha256` to find every other host presenting the same certificate — the strongest cross-asset link for tying acquisitions and shadow infra to the target.
+
+### Naming-Convention Inference
+
+Wildcard SANs and observed hostnames expose the org's naming scheme; generate targeted candidates from it (`<service>.<env>.example.com`) rather than blind brute force.
+
+### IP-First Discovery
+
+For ASN-owned ranges, sweep IPs directly with `naabu`/`httpx` and read served certs (`tlsx`) to find services that have no DNS name at all.
+
+## Advanced Techniques
+
+- **Active SAN harvesting** across whole ranges with `tlsx`/`cero` recovers internal hostnames never logged to public CT.
+- **Favicon and response hashing** (`httpx -favicon`, hash pivots in Shodan) clusters instances of the same app across unrelated hostnames.
+- **Vhost differentials**: probe a single IP with multiple `Host:` values to unmask co-located apps behind one address.
+- **Historical CT/DNS diffing** highlights recently issued certs and newly appearing hosts — high-signal for fresh or misconfigured deployments.
+
+## Consolidation & Probing
+
+1. **Dedupe** names and IPs into one inventory; record source(s) per asset for confidence.
+2. **Live probe** with `httpx`, capturing status/title/tech/server and cert SANs in one pass — each grabbed SAN feeds back as a new seed:
+   `httpx -l hosts.txt -sc -title -server -td -tls-grab -json -o assets.jsonl`
+3. **Classify** assets by function from title/tech/path signals: app, API, marketing, auth, CI/CD, observability, storage, admin, VCS, mail. Cluster by role, not by a specific product.
+4. **Port sweep** interesting hosts with `naabu` for non-HTTP services (DBs, caches, brokers, mgmt ports).
+5. **Prioritize** by exposure and value, then hand each finding to the right specialist skill:
+   - Exposed dashboards / debug / observability / metadata leaks → `information_disclosure`
+   - Login/admin panels with default or weak creds → `weak_password_detection`
+   - Dangling DNS / unclaimed provider resources → `subdomain_takeover`
+   - Cloud consoles/metadata surfaces → `aws` / `gcp` / `kubernetes`
+
+## Testing Methodology
+
+1. **Seed** - domains, org/legal names, known IPs, email domains, code-host org
+2. **Certificate transparency** - pull all logged certs per seed domain and org name (crt.sh, `uncover`)
+3. **SAN/CN extraction** - parse every Subject CN and SAN with `tlsx`; each new name is a new seed
+4. **Passive DNS** - resolve forward and reverse with `dnsx`; harvest historical records
+5. **ASN/IP mapping** - `asnmap` → `mapcidr` to expand owned ranges, then sweep for live hosts
+6. **Active TLS pivot** - `tlsx`/`cero` on live IPs/ports to grab SANs missing from public CT
+7. **Consolidate & probe** - dedupe, `httpx` probe, classify, and route to specialists
+
+## Validation
+
+1. Confirm each discovered asset actually resolves and serves content (live `httpx` result, not just a passive hit)
+2. Attribute assets to the target via matching cert org, shared cert fingerprint, or DNS under a seed domain
+3. Deduplicate vhost aliases and CDN edges down to distinct origins so the surface is not inflated
+4. Record provenance (which source produced each asset) for reproducibility
+
+## False Positives
+
+- CDN/edge hostnames and provider default names that are not org-owned
+- Shared-hosting neighbors on the same IP (vhost co-tenancy, not the target's asset)
+- Stale historical DNS entries pointing at reassigned infrastructure
+- Wildcard-cert-implied hostnames that never actually resolve or serve content
+
+## Impact
+
+- Expanded attack surface: forgotten, staging, and internal-named hosts brute force misses
+- Discovery of misconfigured or unauthenticated services fronted by leaked internal hostnames
+- Attribution of shadow infra, acquisitions, and sibling domains to the target
+- A prioritized, classified inventory that feeds every downstream specialist skill
+
+## Pro Tips
+
+1. Loop the pipeline — every SAN, PTR, and CNAME target is a new seed until the set converges.
+2. crt.sh is the cheapest high-yield source (no key); Censys/Shodan via `uncover` add cert-fingerprint and vhost pivoting when keys exist.
+3. Always cert-grab live hosts with `tlsx` — active SANs catch internal hostnames never sent to public CT.
+4. Internal-looking SANs (`*.internal`, `*.svc.cluster.local`, staging names) are the highest-signal leads.
+5. Wildcard SANs reveal naming conventions — seed targeted guesses instead of blind brute force.
+6. Cluster by function, not product name, so the workflow generalizes to any exposed service.
+7. Keep JSON output throughout so stages chain cleanly (`subfinder` → `dnsx` → `httpx` → `naabu`).
+
+## Summary
+
+Broad passive discovery — CT + TLS SAN pivoting + passive DNS + ASN/IP mapping, looped until convergence — finds the assets brute force misses, especially internal-named and forgotten services leaked through certificates. Build the inventory with the projectdiscovery suite, probe and classify it generically, then route each interesting asset to the specialist skill for its class.
@@ -365,6 +365,23 @@ agent-browser dialog accept "text"    # accept with prompt input
 agent-browser dialog dismiss          # cancel
 ```

+## Readiness & recovery
+
+The first `agent-browser open` in a session launches the headless-Chrome
+daemon; later commands reuse it. Distinguish the two failure modes and react
+differently — do **not** blindly re-run the same failing command in a loop:
+
+- **Daemon / connection failure** (`Failed to connect`, `connection refused`,
+  socket missing, `browser not running`): the daemon isn't up or has died. Run
+  `agent-browser doctor` (add `--fix` if it reports repairable problems), then
+  re-open the page. Retrying the original command unchanged will keep failing.
+- **Malformed command** (`Unknown command`, `Ref not found`, bad flag): fix the
+  command itself — re-snapshot for fresh refs, or correct the syntax.
+
+Invoke `agent-browser` directly through `exec_command`; there is no need to wrap
+it in an extra `sh -c "..."` / `bash -lc "..."` layer, which only adds shell
+quoting and startup-file pitfalls.
+
 ## Diagnosing install issues

 If a command fails unexpectedly (`Unknown command`, `Failed to connect`,
@@ -24,7 +24,15 @@ High-signal flags:
 - `-p, -parallelism <n>` concurrent input targets
 - `-rl, -rate-limit <n>` request rate limit
 - `-timeout <seconds>` request timeout
+- `-ct, -crawl-duration <s|m|h|d>` maximum time to crawl the target
 - `-retry <n>` retry count
+- `-mdp, -max-domain-pages <n>` cap pages crawled per domain (default: unlimited)
+- `-fsu, -filter-similar` collapse similar URLs (e.g. /users/123 and /users/456)
+- `-fs, -field-scope <dn|rdn|fqdn|regex>` crawl scope (default `rdn` = root domain + ALL subdomains)
+- `-f, -field <url|path|...>` emit only one field (e.g. `-f url` for a plain URL list)
+- `-or, -omit-raw` omit raw request/response from JSONL output
+- `-ob, -omit-body` omit response body from JSONL output
+- `-mrs, -max-response-size <bytes>` cap per-response bytes read (default 4194304)
 - `-ef, -extension-filter <list>` extension exclusions
 - `-tlsi, -tls-impersonate` experimental JA3/TLS impersonation
 - `-hl, -headless` enable hybrid headless crawling
@@ -37,13 +45,13 @@ High-signal flags:
 - `-silent`, `-j, -jsonl`, `-o <file>` output controls

 Agent-safe baseline for automation:
-`mkdir -p crawl && katana -u https://target.tld -d 3 -jc -kf robotstxt -c 10 -p 10 -rl 50 -timeout 10 -retry 1 -ef png,jpg,jpeg,gif,svg,css,woff,woff2,ttf,eot,map -silent -j -o crawl/katana.jsonl`
+`mkdir -p crawl && katana -u https://target.tld -d 3 -ct 10m -mdp 2000 -fsu -jc -kf robotstxt -c 10 -p 10 -rl 50 -timeout 10 -retry 1 -ef png,jpg,jpeg,gif,svg,css,woff,woff2,ttf,eot,map -silent -j -o crawl/katana.jsonl`

 Common patterns:
 - Fast crawl baseline:
  `katana -u https://target.tld -d 3 -jc -silent`
- Deeper JS-aware crawl:
-  `katana -u https://target.tld -d 5 -jc -jsl -kf all -c 10 -p 10 -rl 50 -o katana_urls.txt`
+- Deeper JS-aware crawl (narrowed target; keep it time-bounded):
+  `katana -u https://target.tld -d 5 -ct 15m -jc -jsl -kf all -c 10 -p 10 -rl 50 -o katana_urls.txt`
 - Multi-target run with JSONL output:
  `katana -list urls.txt -d 3 -jc -silent -j -o katana.jsonl`
 - Headless crawl with local Chrome:
@@ -59,6 +67,13 @@ Critical correctness rules:
 - For `-kf`, keep depth at least `-d 3` so known files are fully covered.
 - If writing to a file, ensure parent directory exists before `-o`.

+Keeping output small (katana has NO default page cap, so plan for volume):
+- Bound scope and volume: `-fs fqdn` (or `-cs`/`-cos` regex) so the crawl doesn't wander across every subdomain, `-mdp <n>` to cap pages per domain, `-fsu` to collapse near-identical URLs, and `-ct`/`-d` to bound time and depth.
+- Shrink each record: default JSONL is verbose. If you only need endpoints, emit a plain URL list with `-f url` instead of `-j`. If you need JSONL, drop the heavy parts with `-or` (omit raw) and `-ob` (omit body), and lower `-mrs` to cap per-response bytes.
+- Reserve `-jsl` / `-kf all` / higher `-d` for a specific narrowed target — they multiply output fast on large sites.
+- Reduce, then delete: once the crawl finishes, extract just what you need (e.g. `katana ... -f url -o urls.txt` or `sort -u` a URL list, or a short note of interesting paths) and remove the raw crawl file/dir. Don't keep large raw crawls around after you've distilled them.
+- Sanity-check size (`du -sh <out>`); if it's outsized for the scope, tighten `-fs`/`-mdp`/`-fsu`/`-d`/`-ct` and re-run rather than keeping it.
+
 Usage rules:
 - Keep `-d`, `-c`, `-p`, and `-rl` explicit for reproducible runs.
 - Use `-ef` early to reduce static-file noise before fuzzing.
@@ -7,9 +7,9 @@ description: Run Python through exec_command in the SDK sandbox. Use the image-b

 Use `exec_command` for Python. There is no separate Strix Python executor.

-Prefer writing reusable scripts to `/workspace/scratch/<name>.py` and
-running them with `python3 /workspace/scratch/<name>.py`. For short
-one-off transformations, `python3 -c` or a small here-document is fine.
+Prefer writing reusable scripts to a `.py` file and running them with
+`python3 <name>.py`. For short one-off transformations, `python3 -c` or a
+small here-document is fine.

 The `shell` parameter on `exec_command` is for swapping POSIX shells
 (`bash`/`zsh`/`sh`), not for picking interpreters. Put the interpreter
@@ -84,17 +84,26 @@ automatically, so it shows up in `list_requests` and you can use
 For iterative exploit work, put code in a file:

 ```text
-1. Create or edit `/workspace/scratch/exploit.py` with `apply_patch`.
-2. Run it with `exec_command`: `python3 /workspace/scratch/exploit.py`.
+1. Create or edit a task-unique script (e.g. `poc_<task-id>.py`, so it can't
+   clobber a project file or another agent's script) with `apply_patch`.
+2. Run it with `exec_command`: `python3 poc_<task-id>.py`.
 3. Edit and rerun until the proof-of-concept is reliable.
 ```

 ## Installing extra packages

-The sandbox's Python lives in `/app/.venv`. To add a one-off dependency
-for an exploit script, use `uv` (already in the image and much faster
-than pip):
+The sandbox's Python lives in `/app/.venv`, and it is the active virtualenv
+(`python3` / `pip` already resolve to it). The following common libraries are
+**pre-installed** — import them directly, no install step needed:
+`requests`, `httpx`, `beautifulsoup4` (`bs4`), `lxml`, `pyjwt` (`jwt`),
+`cryptography`.
+
+To add a one-off dependency for an exploit script, use `uv` (already in the
+image and much faster than pip):

 ```bash
 uv pip install --python /app/.venv/bin/python <package>
 ```
+
+Plain `pip install <package>` also works because the venv is active. Install
+before you import, so scripts don't fail with `ModuleNotFoundError`.
@@ -130,6 +130,13 @@ TLS clues: certificate CN/SAN referencing provider default host instead of the c
 3. Optional: issue a DV certificate (legal scope) and reference CT entry as evidence
 4. Demonstrate impact chains (CSP/script-src trust, OAuth redirect acceptance, cookie Domain scoping)

+## Severity
+
+- Score severity based on current claimability plus trusted-origin impact, not just a provider-branded error page
+- When evaluating severity, use `web_search` (if available) for the exact provider/product to confirm whether it now enforces subdomain takeover prevention such as TXT/custom-domain ownership verification or reserved-hostname protections; if search is unavailable, do not treat that absence as evidence that the provider prevents claiming
+- If you have positively confirmed the provider currently prevents third-party claiming and you cannot bypass that control, treat the finding as low severity rather than a confirmed takeover — an unconfirmed provider control is not grounds for downgrading
+- Reserve high/critical severity for cases where you can claim the resource or strongly prove claimability and show meaningful impact such as OAuth redirect abuse, cookie scope abuse, CSP trust, email receipt, or NS delegation control. E.g. Elastic Beanstalk takeovers are still generally legitimate.
+
 ## False Positives

 - "Unknown domain" pages that are not claimable due to enforced TXT/ownership checks
@@ -0,0 +1,212 @@
+---
+name: weak-password-detection
+description: Weak password detection, credential stuffing, and brute-force testing using common passwords, system-generated credentials, and tooling like Hydra
+---
+
+# Weak Password Detection / Credential Brute-Force
+
+Weak or default credentials remain one of the most prevalent and high-impact vulnerabilities. This skill covers systematic detection of weak passwords through dictionary attacks, credential stuffing, system-generated password prediction, and brute-force tooling.
+
+## Attack Surface
+
+- Login portals (web, API, mobile, SSH, FTP, Telnet, RDP)
+- Admin panels, dashboards, and management interfaces
+- Default or hardcoded credentials in applications and devices
+- Self-registration flows with weak password policies
+- Password reset flows that generate predictable tokens or passwords
+- API key and token authentication with weak secrets
+
+## Reconnaissance
+
+### Identify Authentication Endpoints
+
+- Standard login forms: `/login`, `/signin`, `/auth`, `/authenticate`, `/api/login`
+- Admin panels: `/admin`, `/administrator`, `/manage`, `/console`, `/cpanel`
+- API auth: `/api/v1/token`, `/oauth/token`, `/api/auth`, `/graphql` (login mutations)
+- Service ports: SSH (22), FTP (21), Telnet (23), SMB (445), RDP (3389), MySQL (3306), PostgreSQL (5432), Redis (6379), MongoDB (27017)
+- Mobile app login endpoints and deep-link auth handlers
+
+### Determine Authentication Mechanism
+
+- Form-based (POST with username/password fields)
+- Basic Authentication (Base64 `Authorization: Basic ...`)
+- Bearer token / JWT (password grant flow)
+- API key in header, query parameter, or body
+- Multi-step authentication (username first, then password)
+- CAPTCHA presence and type (reCAPTCHA, hCaptcha, image-based, math)
+- Rate limiting indicators (429 responses, lockout messages, delays)
+
+### Enumerate Valid Usernames
+
+- Error message differentiation: "Invalid username" vs "Invalid password"
+- Registration page username availability checks
+- Password reset flow: response timing or message leakage
+- Public profiles, API responses, or metadata exposing usernames
+- Common patterns: `admin`, `administrator`, `root`, `user`, `test`, `guest`, `support`, `service`, `api`, `dev`, `ops`
+- Email format derivation from company domain patterns
+
+## Key Vulnerabilities
+
+### Weak Password Policies
+
+- No minimum length or complexity requirements
+- Allowing common passwords: `password`, `123456`, `qwerty`, `admin`, `letmein`
+- Not checking against breached password databases (Have I Been Pwned)
+- Case-insensitive password storage
+- No password history enforcement
+- Excessively short maximum length (indicates plaintext or weak hashing)
+
+### Default and Hardcoded Credentials
+
+- Vendor defaults: `admin/admin`, `admin/password`, `root/root`, `guest/guest`
+- Application frameworks: `django/admin`, `tomcat/tomcat`, `weblogic/weblogic`
+- IoT devices, routers, cameras: manufacturer-specific defaults
+- Database defaults: `postgres/postgres`, `sa/sa`, `root/(empty)`
+- Cloud defaults: AWS instance metadata, Azure default service principals
+- Hardcoded in source code, configuration files, or documentation
+
+### Credential Stuffing
+
+- Users reuse passwords across services
+- Breached credential lists (COMB, Collection #1-5, etc.) enable mass account takeover
+- No multi-factor authentication allows direct access with valid credentials
+- Missing breach detection or forced password rotation after known leaks
+
+### Predictable System-Generated Passwords
+
+- Sequential or pattern-based: `Password1`, `Welcome2025!`, `CompanyName123`
+- Time-based generation: passwords derived from registration timestamp
+- Weak randomness: predictable PRNG seeds in password generators
+- Reset tokens that double as temporary passwords with short expiration
+
+### Brute-Force Vulnerabilities
+
+- No rate limiting on login attempts
+- Absent or ineffective account lockout (client-side only, easily bypassed)
+- IP-based blocking without session/user correlation (rotate IPs via proxy)
+- CAPTCHA bypassable or only triggered after excessive attempts
+- Parallel login attempts not tracked (race conditions on attempt counters)
+- Verbose error messages revealing valid usernames
+
+## Advanced Techniques
+
+### Targeted Password Lists
+
+- Generate custom wordlists from:
+  - Company name, product names, and domain components
+  - Geographic location, industry terms
+  - Season + year patterns: `Summer2025!`, `Winter2026@`
+  - Keyboard walks and leet speak variations
+  - Previously breached passwords for the target domain
+- Cewl: `cewl -d 3 -m 5 -w custom.txt https://target.com` to generate from website content
+
+### Credential Stuffing Workflows
+
+- Use breach databases filtered by target domain or related domains
+- Test email:password pairs where email matches target domain
+- Test username:password pairs with common username derivations
+- Validate successful logins without triggering MFA by checking session endpoints
+
+### Multi-Step Authentication Bypass
+
+- Username enumeration → password brute-force on second step
+- Session fixation between steps: manipulate step identifiers
+- Skip steps via direct URL access to later stages
+- Response manipulation to bypass verification checks
+
+### API and Mobile-Specific
+
+- GraphQL login mutations: batch brute-force via array inputs
+- Mobile APIs often lack rate limiting compared to web frontends
+- JWT password grant flows: brute-force against `/token` endpoint
+- OAuth2 password grant: test `grant_type=password` with weak credentials
+
+### Service-Level Brute-Force
+
+- SSH: `hydra -l admin -P passwords.txt ssh://target.com`
+- FTP: `hydra -L users.txt -P passwords.txt ftp://target.com`
+- RDP: `hydra -l administrator -P passwords.txt rdp://target.com`
+- SMB: `hydra -L users.txt -P passwords.txt smb://target.com`
+- Database: MySQL, PostgreSQL, MongoDB, Redis with weak credentials
+- API endpoints: `ffuf` or custom scripts for HTTP-based brute-force
+
+## Tooling
+
+### Hydra (Primary Tool)
+
+- HTTP POST form brute-force:
+  `hydra -l admin -P /usr/share/wordlists/rockyou.txt target.com http-post-form "/login:username=^USER^&password=^PASS^:Invalid credentials"`
+- Basic Auth:
+  `hydra -L users.txt -P passwords.txt target.com http-get -s 8080 /admin`
+- SSH:
+  `hydra -l root -P passwords.txt -t 4 ssh://target.com`
+- FTP:
+  `hydra -L users.txt -P passwords.txt ftp://target.com`
+- Custom headers and cookies:
+  `hydra ... http-post-form "/api/login:json={\"user\":\"^USER^\",\"pass\":\"^PASS^\"}:F=401"`
+
+### ffuf (HTTP Fuzzing)
+
+- Login brute-force with multiple users and passwords:
+  `ffuf -w users.txt:USER -w passwords.txt:PASS -u https://target.com/login -X POST -d "username=USER&password=PASS" -fr "Invalid"`
+- Filter by response size, status code, or regex to identify successes
+
+### Patator (Versatile Brute-Force)
+
+- `patator http_fuzz url=https://target.com/login method=POST body='username=FILE0&password=FILE1' 0=user.txt 1=pass.txt -x ignore:fgrep='Invalid'`
+
+### Custom Python Scripts
+
+- Use `requests` with threading for high-speed API brute-force
+- Implement jitter and proxy rotation to evade rate limiting
+- Parse CSRF tokens dynamically between requests
+
+### Wordlists
+
+- `/usr/share/wordlists/rockyou.txt` (common passwords)
+- `/usr/share/seclists/Passwords/` (organized by category)
+- `/usr/share/seclists/Passwords/Default-Credentials/` (vendor defaults)
+- Custom lists from Cewl, CeWL, or target-specific scraping
+- Breach compilation subsets filtered by target relevance
+
+## Validation
+
+1. Confirm successful login with captured credentials (session token, cookie, or JWT)
+2. Verify account access level: admin vs user privileges
+3. Check if MFA is enforced post-login or can be bypassed
+4. Test credential reuse across other endpoints or services
+5. Document password policy weaknesses that allowed the breach
+6. Verify if the same credentials work on staging, dev, or related domains
+
+## False Positives
+
+- Honey accounts or honeypot responses designed to mislead attackers
+- Temporary lockouts that resolve quickly (distinguish from permanent bans)
+- Different error messages that don't actually indicate valid username enumeration
+- CAPTCHA or WAF blocking that appears as a failed login
+- Rate limiting that returns 429 instead of 401 (adjust timing)
+
+## Impact
+
+- Complete account takeover for affected users
+- Administrative access leading to full system compromise
+- Lateral movement via reused credentials across services
+- Data exfiltration, privilege escalation, and persistence
+- Reputational damage and compliance violations (GDPR, PCI-DSS)
+
+## Pro Tips
+
+1. Always start with default credentials and vendor-specific lists before broad brute-force
+2. Enumerate usernames first; password brute-force without valid users is inefficient
+3. Use small, targeted wordlists before massive lists like rockyou.txt
+4. Monitor for rate limiting and adapt delays; aggressive brute-force causes IP bans and alerts
+5. Test for password spraying (one password, many users) before targeted brute-force
+6. Check for concurrent session limits; successful logins may kick out legitimate users
+7. GraphQL batching can test multiple credentials in a single request, bypassing per-request limits
+8. Document the password policy and recommend minimum standards (length, complexity, breach checking)
+9. When Hydra is unavailable, use ffuf or custom scripts with equivalent logic
+10. Combine with MFA testing: weak passwords plus missing MFA is a critical finding
+
+## Summary
+
+Weak password detection requires systematic enumeration of authentication surfaces, intelligent wordlist selection, and careful brute-force execution. The highest impact often comes from default credentials, password spraying, and credential stuffing rather than exhaustive brute-force. Always validate findings with confirmed logins and assess the full scope of account compromise.
@@ -2,7 +2,7 @@

 To help make Strix better for everyone, we collect anonymized data that helps us understand how to better improve our AI security agent for our users, guide the addition of new features, and fix common errors and bugs. This feedback loop is crucial for improving Strix's capabilities and user experience.

-We use [PostHog](https://posthog.com), an open-source analytics platform, for data collection and analysis. Our telemetry implementation is fully transparent - you can review the [source code](https://github.com/usestrix/strix/blob/main/strix/telemetry/posthog.py) to see exactly what we track.
+We use [PostHog](https://posthog.com), an open-source analytics platform, for data collection and analysis, along with [Scarf](https://scarf.sh). Our telemetry implementation is fully transparent - you can review the source code ([posthog.py](https://github.com/usestrix/strix/blob/main/strix/telemetry/posthog.py), [scarf.py](https://github.com/usestrix/strix/blob/main/strix/telemetry/scarf.py)) to see exactly what we track.

 ### Telemetry Policy

@@ -16,7 +16,8 @@ We collect only very **basic** usage data including:
 **System Context:** OS type, architecture, Strix version\
 **Scan Context:** Scan mode (quick/standard/deep), scan type (whitebox/blackbox)\
 **Model Usage:** Which LLM model is being used (not prompts or responses)\
-**Aggregate Metrics:** Vulnerability counts by severity
+**Feature Usage:** Which built-in skills are loaded\
+**Aggregate Metrics:** Vulnerability counts by severity and weakness category (CWE)

 ### What We **Never** Collect

@@ -26,10 +26,10 @@ def _is_enabled() -> bool:
    return load_settings().telemetry.enabled


-def _send(event: str, properties: dict[str, Any]) -> None:
+def _send(event: str, properties: dict[str, Any]) -> bool:
    if not _is_enabled():
        logger.debug("posthog disabled; skipping event %s", event)
-        return
+        return False
    try:
        payload = {
            "api_key": _POSTHOG_PUBLIC_API_KEY,
@@ -46,8 +46,10 @@ def _send(event: str, properties: dict[str, Any]) -> None:
            pass
    except Exception:  # noqa: BLE001
        logger.debug("posthog send failed for event %s", event, exc_info=True)
+        return False
    else:
        logger.debug("posthog event sent: %s", event)
+        return True


 def start(
@@ -71,17 +73,34 @@ def start(
    )


-def finding(severity: str) -> None:
+def finding(severity: str, cwe: str | None = None, is_cve: bool = False) -> None:
    _send(
        "finding_reported",
        {
            **base_props(),
            "severity": severity.lower(),
+            "cwe": (cwe or "").strip().lower() or "unknown",
+            "is_cve": is_cve,
+        },
+    )
+
+
+def skill_loaded(skill_name: str) -> None:
+    _send(
+        "skill_loaded",
+        {
+            **base_props(),
+            "skill": skill_name,
        },
    )


 def end(report_state: "ReportState", exit_reason: str = "completed") -> None:
+    if report_state.posthog_scan_ended_sent:
+        return
+    if report_state.scan_ended_exit_reason is None:
+        report_state.scan_ended_exit_reason = exit_reason
+
    vulnerabilities_counts = {"critical": 0, "high": 0, "medium": 0, "low": 0, "info": 0}
    for v in report_state.vulnerability_reports:
        sev = v.get("severity", "info").lower()
@@ -110,11 +129,11 @@ def end(report_state: "ReportState", exit_reason: str = "completed") -> None:
    except (TypeError, ValueError, AttributeError):
        pass

-    _send(
+    report_state.posthog_scan_ended_sent = _send(
        "scan_ended",
        {
            **base_props(),
-            "exit_reason": exit_reason,
+            "exit_reason": report_state.scan_ended_exit_reason,
            "duration_seconds": round(duration),
            "vulnerabilities_total": len(report_state.vulnerability_reports),
            **{f"vulnerabilities_{k}": v for k, v in vulnerabilities_counts.items()},
@@ -123,6 +142,44 @@ def end(report_state: "ReportState", exit_reason: str = "completed") -> None:
    )


+def viewer_opened(source: str, live: bool) -> None:
+    _send(
+        "viewer_opened",
+        {
+            **base_props(),
+            "source": source,
+            "live": live,
+        },
+    )
+
+
+def viewer_cta_clicked(cta: str, surface: str | None = None) -> None:
+    props = {
+        **base_props(),
+        "cta": cta[:64],
+    }
+    if surface:
+        props["surface"] = surface[:64]
+    _send("viewer_cta_clicked", props)
+
+
+_VIEWER_EMAIL_STEPS = frozenset(
+    {"email_submitted", "email_verified", "report_sent", "work_email_required"}
+)
+
+
+def viewer_email_event(step: str, purpose: str | None = None) -> None:
+    if step not in _VIEWER_EMAIL_STEPS:
+        return
+    _send(
+        f"viewer_{step}",
+        {
+            **base_props(),
+            **({"purpose": purpose} if purpose else {}),
+        },
+    )
+
+
 def error(error_type: str) -> None:
    props = {**base_props(), "error_type": error_type}
    _send("error", props)
@@ -28,10 +28,10 @@ def _is_enabled() -> bool:
    return load_settings().telemetry.enabled


-def _send(event: str, properties: dict[str, Any]) -> None:
+def _send(event: str, properties: dict[str, Any]) -> bool:
    if not _is_enabled():
        logger.debug("scarf disabled; skipping event %s", event)
-        return
+        return False
    try:
        props = dict(properties)
        version = str(props.pop("strix_version", get_version()) or "unknown")
@@ -47,8 +47,10 @@ def _send(event: str, properties: dict[str, Any]) -> None:
            pass
    except Exception:  # noqa: BLE001
        logger.debug("scarf send failed for event %s", event, exc_info=True)
+        return False
    else:
        logger.debug("scarf event sent: %s", event)
+        return True


 def start(
@@ -73,18 +75,36 @@ def start(
    )


-def finding(severity: str) -> None:
+def finding(severity: str, cwe: str | None = None, is_cve: bool = False) -> None:
    _send(
        "finding_reported",
        {
            **base_props(),
            "session": SESSION_ID,
            "severity": severity.lower(),
+            "cwe": (cwe or "").strip().lower() or "unknown",
+            "is_cve": is_cve,
+        },
+    )
+
+
+def skill_loaded(skill_name: str) -> None:
+    _send(
+        "skill_loaded",
+        {
+            **base_props(),
+            "session": SESSION_ID,
+            "skill": skill_name,
        },
    )


 def end(report_state: ReportState, exit_reason: str = "completed") -> None:
+    if report_state.scarf_scan_ended_sent:
+        return
+    if report_state.scan_ended_exit_reason is None:
+        report_state.scan_ended_exit_reason = exit_reason
+
    vulnerabilities_counts = {"critical": 0, "high": 0, "medium": 0, "low": 0, "info": 0}
    for v in report_state.vulnerability_reports:
        sev = v.get("severity", "info").lower()
@@ -115,12 +135,12 @@ def end(report_state: ReportState, exit_reason: str = "completed") -> None:
    except (TypeError, ValueError, AttributeError):
        pass

-    _send(
+    report_state.scarf_scan_ended_sent = _send(
        "scan_ended",
        {
            **base_props(),
            "session": SESSION_ID,
-            "exit_reason": exit_reason,
+            "exit_reason": report_state.scan_ended_exit_reason,
            "duration_seconds": round(duration),
            "vulnerabilities_total": len(report_state.vulnerability_reports),
            **{f"vulnerabilities_{k}": v for k, v in vulnerabilities_counts.items()},
@@ -229,7 +229,8 @@ async def wait_for_message(  # noqa: PLR0911
    Use when you have nothing useful to do until a child/peer responds
    — typically after spawning subagents and you want to wait for
    their completion reports. The agent automatically resumes when any
-    message arrives.
+    message arrives, so pick a ``timeout_seconds`` proportional to the
+    work you're awaiting.

    **Critical caveats:**

@@ -246,9 +247,19 @@ async def wait_for_message(  # noqa: PLR0911
        reason: One-line note shown in graph snapshots while you're
            waiting (helps a human or sibling agent debug who's stuck
            on what).
-        timeout_seconds: Hard cap (default 600s). On timeout the tool
-            returns and you decide whether to keep working or wait
-            again.
+        timeout_seconds: Max seconds to wait (default 600). This is only
+            a cap — the tool returns the INSTANT a message arrives, so a
+            larger value never makes you wait longer when the reply does
+            come. Right-size it to what you're waiting on: a short wait
+            (e.g. 10-60s) for a quick ack or a small/fast subtask, and a
+            longer one (e.g. ~100-200s) only for genuinely long-running
+            work (deep recon, exploitation, a full sub-scan). The cap only
+            bites when the expected message never arrives — so an oversized
+            timeout on a trivial wait just strands you idle until it
+            elapses. On timeout the tool returns and you decide whether to
+            keep working or wait again. (Applies to autonomous multi-agent
+            runs; in interactive/chat sessions the agent instead parks until
+            a message arrives and this cap is not enforced.)
    """
    inner = _ctx(ctx)
    coordinator = coordinator_from_context(inner)
@@ -481,9 +492,10 @@ async def agent_finish(
    3. Stops this subagent's execution.

    **Vulnerability findings must already be filed via
-    ``create_vulnerability_report`` before calling this.** The
-    ``findings`` field here is for narrative summary only — it does
-    not register vulns in the scan report.
+    ``create_vulnerability_report`` (or ``create_dependency_report``
+    for known-CVE dependency/supply-chain findings) before calling
+    this.** The ``findings`` field here is for narrative summary only
+    — it does not register vulns in the scan report.

    Write the summary as if the parent has no idea what you were
    doing: what did you test, what did you find/confirm/rule out,
@@ -494,8 +506,9 @@ async def agent_finish(
            and specific (URLs, parameters, payloads that worked).
        findings: Optional bullet list of confirmed observations. For
            credit-bearing vulnerabilities, file
-            ``create_vulnerability_report`` first; this is for
-            narrative.
+            ``create_vulnerability_report`` first (or
+            ``create_dependency_report`` for dependency CVEs); this is
+            for narrative.
        success: Whether the assigned subtask was completed
            successfully. Default ``True``.
        report_to_parent: Whether to deliver the completion report to
@@ -96,6 +96,15 @@ async def finish_scan(
    2. Writes the four narrative sections to the scan record.
    3. Marks the scan completed and stops execution.

+    **This is a terminal action, not a status probe.** Whatever you pass
+    is persisted VERBATIM as the final, customer-facing report and then
+    execution stops. There is no draft mode and no second chance: never
+    submit placeholder, provisional, or "checking if done" text in any
+    field, and never call ``finish_scan`` to poll whether subagents are
+    done (use ``view_agent_graph`` / ``wait_for_message`` for that).
+    Call it exactly ONCE, only when every field holds genuine, finished
+    assessment prose.
+
    **Pre-flight checklist (mandatory — do not skip):**

    1. **Call ``view_agent_graph`` first.** Inspect every entry in the
@@ -108,19 +117,39 @@ async def finish_scan(
       Calling ``finish_scan`` while children are alive orphans their
       work and produces an incomplete report.
    2. All vulnerabilities you found are filed via
-       ``create_vulnerability_report`` (un-reported findings are not
-       tracked and not credited).
+       ``create_vulnerability_report`` — or, for known-CVE dependency
+       findings, ``create_dependency_report`` (un-reported findings are
+       not tracked and not credited). A dependency CVE already filed via
+       ``create_dependency_report`` counts as reported; it does NOT need
+       re-filing here and does NOT block finishing.
    3. Don't double-report — one report per distinct vulnerability.
+    4. **Attack-chaining gate.** Do NOT finish until you have genuinely
+       considered chaining the confirmed findings into higher-impact,
+       end-to-end attack paths and tested every plausibly-related
+       combination. You may rule out combinations you can confidently
+       call unrelated — note why instead of padding chains. Any
+       validated chain must already be filed via
+       ``create_vulnerability_report`` — a demonstrated end-to-end chain
+       is a PoC-backed vulnerability, so it uses that tool even when one
+       link is a dependency CVE (the standalone CVE stays in its own
+       ``create_dependency_report``) — and surfaced prominently in
+       ``executive_summary`` / ``technical_analysis``. Finding no real
+       chain after a serious attempt is acceptable; skipping the
+       chaining reasoning, or ignoring a plausibly-related combination,
+       is not.

    **Calling this multiple times overwrites the previous report.**
    Make the single call comprehensive.

-    **Customer-facing report rules** (this output is rendered into the
-    final PDF the client sees):
+    **Report output rules** (this content may be rendered into generated
+    reports):

    - Never mention internal infrastructure: no local/absolute paths
      (``/workspace/...``), no agent names, no sandbox/orchestrator/
      tooling references, no system prompts, no model-internal errors.
+      Never leak internal identifiers (proxy request IDs, internal
+      vulnerability report IDs, or any system-generated IDs) into any
+      field.
    - Tone: formal, third-person, objective, concise. This is a
      consultant deliverable, not an engineering log.
    - Each section has a specific role:
@@ -140,6 +169,74 @@ async def finish_scan(
          (Immediate / Short-term / Medium-term), each with concrete
          remediation steps. End with retest/validation guidance.

+    - **Formatting — use markdown in every field.** These fields may be
+      rendered into generated reports, so structure them clearly: lead
+      each section with a short ``# Heading``, use ``**bold**`` for labels/emphasis,
+      ``inline code`` for identifiers/paths/parameters, bullet or
+      numbered lists for enumerations, and fenced code blocks
+      (```` ```language ````) for any code/payload excerpts. Never emit
+      one flat wall of prose or leave code unformatted.
+    - If **zero** vulnerabilities were found, say so plainly and
+      characterize the posture positively; ``technical_analysis`` should
+      summarize the areas tested and confirm no issues, and
+      ``recommendations`` should focus on general hardening.
+
+    Example (abbreviated — mirror this structure, not the wording)::
+
+        executive_summary:
+            # Executive Summary
+
+            An external assessment of the **Acme Customer Portal**
+            identified multiple weaknesses that could lead to
+            unauthorized access to customer data.
+
+            **Overall risk posture:** Elevated.
+
+            **Key findings**
+            - Confirmed SSRF in a URL-preview feature reaching internal
+              network ranges.
+            - Broken tenant isolation enabling cross-tenant data access.
+
+            **Business impact**
+            - Potential exposure of customer records across tenants.
+
+        methodology:
+            # Methodology
+
+            Conducted per the **OWASP WSTG**.
+
+            **Engagement type:** Gray-box external test.
+            **Scope:** `https://app.acme.example`, `.../api/v1/`.
+
+            **Activities:** recon, authn/session review, authorization
+            and tenant-isolation testing, input/SSRF testing.
+
+        technical_analysis:
+            # Technical Analysis
+
+            **Severity model** reflects exploitability x impact.
+
+            1. **SSRF in URL preview** (Critical) — insufficient
+               destination validation; reaches link-local addresses.
+            2. **Broken tenant isolation** (High) — object identifiers
+               accepted without ownership checks.
+
+            **Systemic themes:** authorization enforced inconsistently;
+            no deny-by-default egress policy.
+
+        recommendations:
+            # Recommendations
+
+            **Immediate**
+            1. Remediate SSRF: enforce a destination allowlist,
+               deny-by-default, re-validate on every redirect hop.
+
+            **Short-term**
+            2. Centralize authorization with deny-by-default middleware.
+
+            **Retest & validation:** re-test immediate items to confirm
+            SSRF and tenant-isolation controls hold.
+
    Args:
        executive_summary: Business-level summary for leadership.
        methodology: Frameworks, scope, and approach.
@@ -21,6 +21,8 @@ from caido_sdk_client.types import (


 if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable
+
    from caido_sdk_client import Client as CaidoClient


@@ -42,6 +44,7 @@ _SITEMAP_PAGE_SIZE = 30

 _DEFAULT_CAIDO_URL = "http://127.0.0.1:48080"
 _CLIENT_CACHE: dict[str, Client] = {}
+_CLIENT_LOCK = asyncio.Lock()
 _REQ_FIELD_MAP: dict[SortBy, tuple[str, str]] = {
    "timestamp": ("req", "created_at"),
    "host": ("req", "host"),
@@ -81,19 +84,46 @@ def _login_as_guest() -> str:
    return str(payload["data"]["loginAsGuest"]["token"]["accessToken"])


-async def get_client() -> Client:
-    if client := _CLIENT_CACHE.get("default"):
-        return client
-
+async def _new_client() -> Client:
    token = await asyncio.to_thread(_login_as_guest)
    client = Client(caido_url(), auth=TokenAuthOptions(token=token))
    await client.connect()
-    _CLIENT_CACHE["default"] = client
    return client


+async def get_client() -> Client:
+    """Return the shared Caido client, creating it under a lock if needed.
+
+    The lock prevents two concurrent callers from each building a client and
+    racing ``connect()`` on the same transport ("Transport is already
+    connected").
+    """
+    async with _CLIENT_LOCK:
+        client = _CLIENT_CACHE.get("default")
+        if client is None:
+            client = await _new_client()
+            _CLIENT_CACHE["default"] = client
+        return client
+
+
+async def call_with_client[T](fn: Callable[[Client], Awaitable[T]]) -> T:
+    """Run ``fn`` against the shared client, serialized through ``_CLIENT_LOCK``.
+
+    The Caido GraphQL transport is not safe for concurrent use: two in-flight
+    requests race and raise "Transport is already connected". Serializing every
+    proxy call through the lock prevents that.
+    """
+    async with _CLIENT_LOCK:
+        client = _CLIENT_CACHE.get("default")
+        if client is None:
+            client = await _new_client()
+            _CLIENT_CACHE["default"] = client
+        return await fn(client)
+
+
 async def close_client() -> None:
-    client = _CLIENT_CACHE.pop("default", None)
+    async with _CLIENT_LOCK:
+        client = _CLIENT_CACHE.pop("default", None)
    if client is None:
        return
    await client.aclose()
@@ -385,19 +415,23 @@ async def list_requests(
    sort_order: SortOrder = "desc",
    scope_id: str | None = None,
 ) -> Any:
-    return await list_requests_with_client(
-        await get_client(),
-        httpql_filter=httpql_filter,
-        first=first,
-        after=after,
-        sort_by=sort_by,
-        sort_order=sort_order,
-        scope_id=scope_id,
+    return await call_with_client(
+        lambda client: list_requests_with_client(
+            client,
+            httpql_filter=httpql_filter,
+            first=first,
+            after=after,
+            sort_by=sort_by,
+            sort_order=sort_order,
+            scope_id=scope_id,
+        )
    )


 async def view_request(request_id: str, *, part: RequestPart = "request") -> Any:
-    return await get_request_with_client(await get_client(), request_id, part=part)
+    return await call_with_client(
+        lambda client: get_request_with_client(client, request_id, part=part)
+    )


 async def repeat_request(
@@ -406,22 +440,26 @@ async def repeat_request(
    modifications: dict[str, Any] | None = None,
 ) -> dict[str, Any]:
    mods = modifications or {}
-    result = await get_request_with_client(await get_client(), request_id, part="request")
-    if result is None or result.request.raw is None:
-        raise ValueError(f"Request {request_id} not found")

-    original = result.request
-    raw_str = result.request.raw.decode("utf-8", errors="replace")
-    components = parse_raw_request(raw_str)
-    full_url = full_url_from_components(original, components, mods)
-    modified = apply_modifications(components, mods, full_url)
-    connection, raw = build_raw_request(
-        method=modified["method"],
-        url=modified["url"],
-        headers=modified["headers"],
-        body=modified["body"],
-    )
-    return await replay_send_raw(await get_client(), raw=raw, connection=connection)
+    async def _run(client: CaidoClient) -> dict[str, Any]:
+        result = await get_request_with_client(client, request_id, part="request")
+        if result is None or result.request.raw is None:
+            raise ValueError(f"Request {request_id} not found")
+
+        original = result.request
+        raw_str = result.request.raw.decode("utf-8", errors="replace")
+        components = parse_raw_request(raw_str)
+        full_url = full_url_from_components(original, components, mods)
+        modified = apply_modifications(components, mods, full_url)
+        connection, raw = build_raw_request(
+            method=modified["method"],
+            url=modified["url"],
+            headers=modified["headers"],
+            body=modified["body"],
+        )
+        return await replay_send_raw(client, raw=raw, connection=connection)
+
+    return await call_with_client(_run)


 async def scope_rules(
@@ -432,7 +470,28 @@ async def scope_rules(
    scope_id: str | None = None,
    scope_name: str | None = None,
 ) -> Any:
-    client = await get_client()
+    async def _run(client: CaidoClient) -> Any:
+        return await _scope_rules_with_client(
+            client,
+            action,
+            allowlist=allowlist,
+            denylist=denylist,
+            scope_id=scope_id,
+            scope_name=scope_name,
+        )
+
+    return await call_with_client(_run)
+
+
+async def _scope_rules_with_client(
+    client: CaidoClient,
+    action: ScopeAction,
+    *,
+    allowlist: list[str] | None = None,
+    denylist: list[str] | None = None,
+    scope_id: str | None = None,
+    scope_name: str | None = None,
+) -> Any:
    if action == "list":
        result = await scope_list(client)
    elif action == "get":
@@ -651,18 +710,20 @@ async def list_sitemap(
    page: int = 1,
    page_size: int = _SITEMAP_PAGE_SIZE,
 ) -> dict[str, Any]:
-    return await list_sitemap_with_client(
-        await get_client(),
-        scope_id=scope_id,
-        parent_id=parent_id,
-        depth=depth,
-        page=page,
-        page_size=page_size,
+    return await call_with_client(
+        lambda client: list_sitemap_with_client(
+            client,
+            scope_id=scope_id,
+            parent_id=parent_id,
+            depth=depth,
+            page=page,
+            page_size=page_size,
+        )
    )


 async def view_sitemap_entry(entry_id: str) -> dict[str, Any]:
-    return await view_sitemap_entry_with_client(await get_client(), entry_id)
+    return await call_with_client(lambda client: view_sitemap_entry_with_client(client, entry_id))


 __all__ = [
@@ -2,6 +2,7 @@

 from __future__ import annotations

+import asyncio
 import dataclasses
 import json
 import logging
@@ -19,6 +20,8 @@ logger = logging.getLogger(__name__)


 if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable
+
    from caido_sdk_client import Client

    from strix.tools.proxy.caido_api import (
@@ -38,12 +41,23 @@ else:

 ScopeAction = Literal["get", "list", "create", "update", "delete"]

+# All agents in a scan share one host-side Caido client whose GraphQL transport
+# is not concurrency-safe (parallel calls raise "Transport is already
+# connected"). Serialize every host-side proxy call through this lock.
+_CAIDO_CALL_LOCK = asyncio.Lock()
+

 def _ctx_client(ctx: RunContextWrapper) -> Client | None:
    inner = ctx.context if isinstance(ctx.context, dict) else {}
    return inner.get("caido_client")


+async def _call[T](client: Client, fn: Callable[[Client], Awaitable[T]]) -> T:
+    """Run ``fn`` against the shared client, serialized under ``_CAIDO_CALL_LOCK``."""
+    async with _CAIDO_CALL_LOCK:
+        return await fn(client)
+
+
 def _to_tool_json(value: Any) -> Any:
    """Recursively convert SDK dataclasses/Pydantic objects to tool JSON values."""
    if value is None or isinstance(value, str | int | float | bool):
@@ -146,14 +160,17 @@ async def list_requests(
        return _no_client()

    try:
-        connection = await caido_api.list_requests_with_client(
+        connection = await _call(
            client,
-            httpql_filter=httpql_filter,
-            first=first,
-            after=after,
-            sort_by=sort_by,
-            sort_order=sort_order,
-            scope_id=scope_id,
+            lambda client: caido_api.list_requests_with_client(
+                client,
+                httpql_filter=httpql_filter,
+                first=first,
+                after=after,
+                sort_by=sort_by,
+                sort_order=sort_order,
+                scope_id=scope_id,
+            ),
        )

        entries = []
@@ -249,7 +266,10 @@ async def view_request(
        return _no_client()

    try:
-        result = await caido_api.get_request_with_client(client, request_id, part=part)
+        result = await _call(
+            client,
+            lambda client: caido_api.get_request_with_client(client, request_id, part=part),
+        )
        if result is None:
            return json.dumps(
                {"success": False, "error": f"Request {request_id} not found"},
@@ -364,15 +384,10 @@ async def repeat_request(
        return _no_client()
    mods = modifications or {}

-    try:
+    async def _do(client: Client) -> dict[str, Any] | None:
        result = await caido_api.get_request_with_client(client, request_id, part="request")
        if result is None or result.request.raw is None:
-            return json.dumps(
-                {"success": False, "error": f"Request {request_id} not found"},
-                ensure_ascii=False,
-                default=str,
-            )
-
+            return None
        original = result.request
        raw_str = result.request.raw.decode("utf-8", errors="replace")
        components = caido_api.parse_raw_request(raw_str)
@@ -384,7 +399,16 @@ async def repeat_request(
            headers=modified["headers"],
            body=modified["body"],
        )
-        replay = await caido_api.replay_send_raw(client, raw=raw, connection=connection)
+        return await caido_api.replay_send_raw(client, raw=raw, connection=connection)
+
+    try:
+        replay = await _call(client, _do)
+        if replay is None:
+            return json.dumps(
+                {"success": False, "error": f"Request {request_id} not found"},
+                ensure_ascii=False,
+                default=str,
+            )
        return _format_replay_tool_result(replay)
    except Exception as exc:  # noqa: BLE001
        return _err("repeat_request", exc)
@@ -441,12 +465,15 @@ async def list_sitemap(
    if client is None:
        return _no_client()
    try:
-        payload = await caido_api.list_sitemap_with_client(
+        payload = await _call(
            client,
-            scope_id=scope_id,
-            parent_id=parent_id,
-            depth=depth,
-            page=page,
+            lambda client: caido_api.list_sitemap_with_client(
+                client,
+                scope_id=scope_id,
+                parent_id=parent_id,
+                depth=depth,
+                page=page,
+            ),
        )
        return json.dumps(payload, ensure_ascii=False, default=str)
    except Exception as exc:  # noqa: BLE001
@@ -472,7 +499,10 @@ async def view_sitemap_entry(
    if client is None:
        return _no_client()
    try:
-        payload = await caido_api.view_sitemap_entry_with_client(client, entry_id)
+        payload = await _call(
+            client,
+            lambda client: caido_api.view_sitemap_entry_with_client(client, entry_id),
+        )
        return json.dumps(payload, ensure_ascii=False, default=str)
    except Exception as exc:  # noqa: BLE001
        return _err("view_sitemap_entry", exc)
@@ -530,7 +560,7 @@ async def scope_rules(

    try:
        if action == "list":
-            scopes = await caido_api.scope_list(client)
+            scopes = await _call(client, caido_api.scope_list)
            return json.dumps(
                {"success": True, "scopes": [_to_tool_json(s) for s in scopes]},
                ensure_ascii=False,
@@ -543,9 +573,11 @@ async def scope_rules(
                    ensure_ascii=False,
                    default=str,
                )
-            scope = await caido_api.scope_get(client, scope_id)
+            scope = await _call(client, lambda client: caido_api.scope_get(client, scope_id))
            return json.dumps(
-                {"success": True, "scope": _to_tool_json(scope)}, ensure_ascii=False, default=str
+                {"success": True, "scope": _to_tool_json(scope)},
+                ensure_ascii=False,
+                default=str,
            )
        if action == "create":
            if not scope_name:
@@ -554,11 +586,16 @@ async def scope_rules(
                    ensure_ascii=False,
                    default=str,
                )
-            scope = await caido_api.scope_create(
-                client, name=scope_name, allowlist=allowlist, denylist=denylist
+            scope = await _call(
+                client,
+                lambda client: caido_api.scope_create(
+                    client, name=scope_name, allowlist=allowlist, denylist=denylist
+                ),
            )
            return json.dumps(
-                {"success": True, "scope": _to_tool_json(scope)}, ensure_ascii=False, default=str
+                {"success": True, "scope": _to_tool_json(scope)},
+                ensure_ascii=False,
+                default=str,
            )
        if action == "update":
            if not scope_id or not scope_name:
@@ -570,11 +607,16 @@ async def scope_rules(
                    ensure_ascii=False,
                    default=str,
                )
-            scope = await caido_api.scope_update(
-                client, scope_id, name=scope_name, allowlist=allowlist, denylist=denylist
+            scope = await _call(
+                client,
+                lambda client: caido_api.scope_update(
+                    client, scope_id, name=scope_name, allowlist=allowlist, denylist=denylist
+                ),
            )
            return json.dumps(
-                {"success": True, "scope": _to_tool_json(scope)}, ensure_ascii=False, default=str
+                {"success": True, "scope": _to_tool_json(scope)},
+                ensure_ascii=False,
+                default=str,
            )
        if not scope_id:
            return json.dumps(
@@ -582,7 +624,7 @@ async def scope_rules(
                ensure_ascii=False,
                default=str,
            )
-        await caido_api.scope_delete(client, scope_id)
+        await _call(client, lambda client: caido_api.scope_delete(client, scope_id))
        return json.dumps(
            {
                "success": True,
@@ -148,8 +148,12 @@ _REQUIRED_FIELDS = {
    "poc_description": "PoC description cannot be empty",
    "poc_script_code": "PoC script/code is REQUIRED - provide the actual exploit/payload",
    "remediation_steps": "Remediation steps cannot be empty",
+    "evidence": "Evidence cannot be empty - provide concrete proof of the finding",
+    "assumptions": "Assumptions cannot be empty - state exploitability prerequisites",
 }

+_VALID_FIX_EFFORT = frozenset({"trivial", "low", "medium", "high"})
+

 async def _do_create(  # noqa: PLR0912
    *,
@@ -161,12 +165,16 @@ async def _do_create(  # noqa: PLR0912
    poc_description: str,
    poc_script_code: str,
    remediation_steps: str,
+    evidence: str,
+    assumptions: str,
+    fix_effort: str,
    cvss_breakdown: dict[str, str],
    endpoint: str | None,
    method: str | None,
    cve: str | None,
    cwe: str | None,
    code_locations: list[dict[str, Any]] | None,
+    fix_pr_body: str | None = None,
    agent_id: str | None = None,
    agent_name: str | None = None,
 ) -> dict[str, Any]:
@@ -180,11 +188,19 @@ async def _do_create(  # noqa: PLR0912
        "poc_description": poc_description,
        "poc_script_code": poc_script_code,
        "remediation_steps": remediation_steps,
+        "evidence": evidence,
+        "assumptions": assumptions,
    }
    for name, msg in _REQUIRED_FIELDS.items():
        if not str(fields.get(name) or "").strip():
            errors.append(msg)

+    fix_effort = (fix_effort or "").strip().lower()
+    if fix_effort not in _VALID_FIX_EFFORT:
+        errors.append(
+            f"Invalid fix_effort: {fix_effort!r}. Must be one of: {sorted(_VALID_FIX_EFFORT)}"
+        )
+
    if not isinstance(cvss_breakdown, dict) or not cvss_breakdown:
        errors.append("cvss_breakdown: must be an object with the 8 CVSS metrics")
        cvss_breakdown = {}
@@ -268,6 +284,9 @@ async def _do_create(  # noqa: PLR0912
            poc_description=poc_description,
            poc_script_code=poc_script_code,
            remediation_steps=remediation_steps,
+            evidence=evidence,
+            assumptions=assumptions,
+            fix_effort=fix_effort,
            cvss=cvss_score,
            cvss_breakdown=cvss_breakdown,
            endpoint=endpoint,
@@ -275,6 +294,7 @@ async def _do_create(  # noqa: PLR0912
            cve=cve,
            cwe=cwe,
            code_locations=parsed_locations,
+            fix_pr_body=fix_pr_body,
            agent_id=agent_id if isinstance(agent_id, str) else None,
            agent_name=agent_name if isinstance(agent_name, str) else None,
        )
@@ -309,12 +329,16 @@ async def create_vulnerability_report(
    poc_description: str,
    poc_script_code: str,
    remediation_steps: str,
+    evidence: str,
+    assumptions: str,
+    fix_effort: str,
    cvss_breakdown: dict[str, str],
    endpoint: str | None = None,
    method: str | None = None,
    cve: str | None = None,
    cwe: str | None = None,
    code_locations: list[dict[str, Any]] | None = None,
+    fix_pr_body: str | None = None,
 ) -> str:
    """File a vulnerability report — one report per fully-verified finding.

@@ -327,24 +351,46 @@ async def create_vulnerability_report(
    - Suspicions you haven't confirmed with a PoC.
    - Tracking multiple vulnerabilities at once — one report per vuln.
    - Re-reporting something you (or another agent) already filed.
+    - Known-CVE dependency / supply-chain findings that can't be
+      dynamically PoC'd — a vulnerable dependency version pinned in a
+      lockfile/manifest that matches a published advisory. File those
+      with ``create_dependency_report`` instead, never with this tool.

    Automatic LLM-based **deduplication** rejects reports that describe
    the same root cause on the same asset as an existing report. If you
    get a ``duplicate_of`` response, do NOT retry — move on to other
    areas.

-    **Customer-facing report rules** (the report is PDF-rendered for
-    delivery):
+    **Report output rules** (this content may be rendered into generated
+    reports):

    - No internal/system details: never mention paths like
      ``/workspace``, internal tools, agents, sandboxes, models, system
      prompts, internal errors / stack traces, or tester environment.
+      Never leak internal identifiers (proxy request IDs, internal
+      report IDs) into any field.
    - Tone: formal, objective, third-person, vendor-neutral, concise.
-    - Standard finding structure: Overview → Severity & CVSS →
-      Affected assets → Technical details → PoC (steps + code) →
-      Impact → Remediation → Evidence (in technical_analysis).
+      Avoid internal-guidance headings like "QUICK", "Approach", or
+      "Techniques" that read like an engineering runbook rather than a
+      client deliverable.
+    - **Use markdown in every text field**: ``**bold**`` for emphasis,
+      ``inline code`` for identifiers/values/parameters, and fenced
+      code blocks (```` ```language ````) for any code/payload/HTTP
+      excerpt. Never leave code bare/unformatted. When referencing a
+      file, annotate the fence, e.g.
+      ```` ```python title=app.py startLineNumber=42 endLineNumber=50 ````.
+    - Field discipline: ``poc_description`` is steps only — NO code (all
+      code goes in ``poc_script_code``); ``remediation_steps`` is prose
+      only — NO code/diffs (code fixes go in ``code_locations``).
    - Numbered steps allowed only in PoC and Remediation sections.
    - Avoid hedging language; be precise and non-vague.
+    - Follow a standard pentest report structure across the fields:
+      (1) overview (``description``), (2) severity & CVSS vector
+      (``cvss_breakdown``), (3) affected asset(s) (``target`` /
+      ``endpoint``), (4) technical details (``technical_analysis``),
+      (5) proof of concept (``poc_description`` + ``poc_script_code``),
+      (6) impact (``impact``), (7) evidence (``evidence``), and
+      (8) remediation (``remediation_steps``).

    **White-box requirement**: when source is available, you MUST
    populate ``code_locations``. See the ``code_locations`` arg below
@@ -407,13 +453,23 @@ async def create_vulnerability_report(
        title: Specific finding title (e.g.
            ``"SQL Injection in /api/users login parameter"``). Don't
            include the CVE number in the title.
-        description: How the vuln was discovered + what it is.
+        description: Concise, non-technical TL;DR of the vulnerability
+            (1-3 sentences) — it appears first in the report. Deep
+            technical detail and root-cause analysis belong in
+            ``technical_analysis``, not here.
        impact: What an attacker achieves; business risk; data at risk.
        target: Affected URL / domain / repository.
        technical_analysis: The mechanism and root cause.
-        poc_description: Step-by-step reproduction.
+        poc_description: Step-by-step reproduction (steps only, no code).
        poc_script_code: Working PoC (Python preferred).
-        remediation_steps: Specific, actionable fix.
+        remediation_steps: Specific, actionable fix (prose, no code).
+        evidence: Concrete proof the issue is real and exploitable —
+            request/response excerpts, observed behavior, tool output.
+            Use fenced code blocks; no internal identifiers/paths.
+        assumptions: Short note on the assumptions/prerequisites that
+            make this finding impactful or exploitable (e.g. "assumes an
+            authenticated low-privilege user").
+        fix_effort: One of ``trivial`` / ``low`` / ``medium`` / ``high``.
        cvss_breakdown: 8-metric object per the format above.
        endpoint: API path / Git path (e.g. ``/api/login``).
        method: HTTP method when relevant.
@@ -482,6 +538,47 @@ async def create_vulnerability_report(
            - Padding ``fix_before`` with surrounding context lines
              that aren't part of the fix.
            - Duplicating the same change across multiple locations.
+        fix_pr_body: Optional. When source is available and you have a
+            concrete fix, a markdown PR-description body proposing the
+            fix (summary + rationale). Prose/markdown only — the code
+            change itself belongs in ``code_locations``. Omit for
+            black-box findings.
+
+    Example (abbreviated — mirror this structure)::
+
+        title: "Reflected XSS in /search q parameter"
+        description:
+            The **`q`** parameter of `/search` reflects user input into
+            the HTML response without encoding, allowing script
+            injection.
+        technical_analysis:
+            The handler interpolates `q` directly into the page body:
+
+            ```python title=views.py startLineNumber=42 endLineNumber=44
+            html = f"<h2>Results for {q}</h2>"
+            return HttpResponse(html)
+            ```
+
+            No output encoding is applied, so `<script>` executes.
+        poc_description:
+            1. Navigate to `/search?q=<payload>`.
+            2. Observe the payload executes in the victim's browser.
+        poc_script_code:
+            ```
+            GET /search?q=<script>alert(document.domain)</script>
+            ```
+        evidence:
+            Response echoes the payload verbatim:
+
+            ```html
+            <h2>Results for <script>alert(document.domain)</script></h2>
+            ```
+        assumptions:
+            Assumes a victim can be induced to open a crafted link.
+        remediation_steps:
+            Context-encode all user input rendered into HTML; prefer the
+            template engine's auto-escaping over string interpolation.
+        fix_effort: "low"
    """
    inner = ctx.context if isinstance(ctx.context, dict) else {}
    raw_agent_id = inner.get("agent_id")
@@ -503,12 +600,327 @@ async def create_vulnerability_report(
        poc_description=poc_description,
        poc_script_code=poc_script_code,
        remediation_steps=remediation_steps,
+        evidence=evidence,
+        assumptions=assumptions,
+        fix_effort=fix_effort,
        cvss_breakdown=cvss_breakdown,
        endpoint=endpoint,
        method=method,
        cve=cve,
        cwe=cwe,
        code_locations=code_locations,
+        fix_pr_body=fix_pr_body,
+        agent_id=agent_id,
+        agent_name=agent_name,
+    )
+    return json.dumps(result, ensure_ascii=False, default=str)
+
+
+_DEP_SEVERITY_FROM_CVSS = {
+    (9.0, 10.0): "critical",
+    (7.0, 9.0): "high",
+    (4.0, 7.0): "medium",
+    (0.0, 4.0): "low",
+}
+
+
+def _dependency_severity(advisory_cvss: float | None) -> tuple[float, str]:
+    if advisory_cvss is None:
+        return 0.0, "info"
+    score = max(0.0, min(10.0, advisory_cvss))
+    for (lo, hi), label in _DEP_SEVERITY_FROM_CVSS.items():
+        if lo <= score < hi or (hi == 10.0 and score == 10.0):
+            return score, label
+    return score, "none"
+
+
+def _build_dependency_metadata(
+    *,
+    package_name: str,
+    installed_version: str,
+    package_ecosystem: str | None,
+    fixed_version: str | None,
+) -> dict[str, str]:
+    metadata = {
+        "package_name": package_name.strip(),
+        "installed_version": installed_version.strip(),
+    }
+    if package_ecosystem and package_ecosystem.strip():
+        metadata["package_ecosystem"] = package_ecosystem.strip()
+    if fixed_version and fixed_version.strip():
+        metadata["fixed_version"] = fixed_version.strip()
+    return metadata
+
+
+def _build_dependency_evidence(
+    *,
+    cve: str,
+    package_name: str,
+    installed_version: str,
+    fixed_version: str | None,
+) -> str:
+    evidence = (
+        f"**Advisory evidence:** `{cve}` applies to `{package_name}` "
+        f"at installed version `{installed_version}`."
+    )
+    if fixed_version and fixed_version.strip():
+        evidence += f" The advisory is fixed in `{fixed_version.strip()}`."
+    return evidence
+
+
+async def _do_create_dependency(  # noqa: PLR0912
+    *,
+    title: str,
+    description: str,
+    target: str,
+    cve: str,
+    package_name: str,
+    installed_version: str,
+    impact: str,
+    remediation_steps: str,
+    assumptions: str,
+    package_ecosystem: str | None,
+    fixed_version: str | None,
+    cwe: str | None,
+    advisory_cvss: float | None,
+    technical_analysis: str | None,
+    fix_effort: str,
+    agent_id: str | None = None,
+    agent_name: str | None = None,
+) -> dict[str, Any]:
+    errors: list[str] = []
+    required = {
+        "title": title,
+        "description": description,
+        "target": target,
+        "package_name": package_name,
+        "installed_version": installed_version,
+        "package_ecosystem": package_ecosystem,
+        "impact": impact,
+        "remediation_steps": remediation_steps,
+        "assumptions": assumptions,
+    }
+    for name, value in required.items():
+        if not str(value or "").strip():
+            errors.append(f"{name} cannot be empty")
+
+    parsed_cve = _extract_cve(cve or "")
+    cve_err = _validate_cve(parsed_cve)
+    if cve_err:
+        errors.append(cve_err)
+
+    if cwe:
+        cwe = _extract_cwe(cwe)
+        cwe_err = _validate_cwe(cwe)
+        if cwe_err:
+            errors.append(cwe_err)
+
+    fix_effort = (fix_effort or "").strip().lower()
+    if fix_effort not in _VALID_FIX_EFFORT:
+        errors.append(
+            f"Invalid fix_effort: {fix_effort!r}. Must be one of: {sorted(_VALID_FIX_EFFORT)}"
+        )
+
+    if advisory_cvss is None:
+        errors.append(
+            "advisory_cvss is required: read the published advisory base score "
+            "(0.0-10.0) off the advisory (trivy CVSS / NVD / GHSA). Severity is "
+            "derived solely from it — do not omit it or the finding cannot be rated."
+        )
+    elif not 0.0 <= advisory_cvss <= 10.0:
+        errors.append(f"advisory_cvss must be between 0.0 and 10.0, got {advisory_cvss}")
+
+    if errors:
+        return {"success": False, "error": "Validation failed", "errors": errors}
+
+    cvss_score, severity = _dependency_severity(advisory_cvss)
+    dependency_metadata = _build_dependency_metadata(
+        package_name=package_name,
+        installed_version=installed_version,
+        package_ecosystem=package_ecosystem,
+        fixed_version=fixed_version,
+    )
+    evidence = _build_dependency_evidence(
+        cve=parsed_cve,
+        package_name=package_name.strip(),
+        installed_version=installed_version.strip(),
+        fixed_version=fixed_version,
+    )
+
+    try:
+        from strix.report.state import get_global_report_state
+
+        report_state = get_global_report_state()
+        if report_state is None:
+            logger.warning("No global report state; dependency report not persisted")
+            return {
+                "success": True,
+                "message": f"Dependency finding '{title}' created (not persisted)",
+                "warning": "Report could not be persisted - report state unavailable",
+            }
+
+        from strix.report.dedupe import check_duplicate
+
+        existing = report_state.get_existing_vulnerabilities()
+        candidate = {
+            "title": title,
+            "description": description,
+            "target": target,
+            "cve": parsed_cve,
+            "dependency_metadata": dependency_metadata,
+            "technical_analysis": technical_analysis,
+        }
+        dedupe = await check_duplicate(candidate, existing)
+        if dedupe.get("is_duplicate"):
+            duplicate_id = dedupe.get("duplicate_id", "")
+            return {
+                "success": False,
+                "error": (
+                    f"Potential duplicate (id={duplicate_id[:8]}...) — "
+                    "do not re-report the same dependency finding"
+                ),
+                "duplicate_of": duplicate_id,
+                "confidence": dedupe.get("confidence", 0.0),
+                "reason": dedupe.get("reason", ""),
+            }
+
+        report_id = report_state.add_vulnerability_report(
+            title=title,
+            description=description,
+            severity=severity,
+            impact=impact,
+            target=target,
+            technical_analysis=technical_analysis,
+            remediation_steps=remediation_steps,
+            evidence=evidence,
+            assumptions=assumptions,
+            fix_effort=fix_effort,
+            cvss=cvss_score if advisory_cvss is not None else None,
+            cve=parsed_cve,
+            cwe=cwe,
+            finding_class="dependency_cve",
+            dependency_metadata=dependency_metadata,
+            agent_id=agent_id if isinstance(agent_id, str) else None,
+            agent_name=agent_name if isinstance(agent_name, str) else None,
+        )
+    except (ImportError, AttributeError) as e:
+        logger.exception("create_dependency_report persistence failed")
+        return {"success": False, "error": f"Failed to create dependency report: {e!s}"}
+    else:
+        logger.info(
+            "Dependency report created: id=%s cve=%s package=%s severity=%s",
+            report_id,
+            parsed_cve,
+            package_name,
+            severity,
+        )
+        return {
+            "success": True,
+            "message": f"Dependency finding '{title}' created successfully",
+            "report_id": report_id,
+            "severity": severity,
+            "cve": parsed_cve,
+        }
+
+
+@function_tool(timeout=180, strict_mode=False)
+async def create_dependency_report(
+    ctx: RunContextWrapper,
+    title: str,
+    description: str,
+    target: str,
+    cve: str,
+    package_name: str,
+    installed_version: str,
+    advisory_cvss: float,
+    impact: str,
+    remediation_steps: str,
+    assumptions: str,
+    package_ecosystem: str,
+    fixed_version: str | None = None,
+    cwe: str | None = None,
+    technical_analysis: str | None = None,
+    fix_effort: str = "low",
+) -> str:
+    """File a known-CVE dependency (SCA) finding — one report per CVE x package.
+
+    Use this instead of ``create_vulnerability_report`` when the finding
+    is a **known-CVE supply-chain issue**: a vulnerable third-party
+    package/version identified from a lockfile, manifest, or SBOM. Unlike
+    a dynamic finding, you do NOT need to trigger the vulnerability with a
+    live PoC — a verified advisory + the affected installed version is the
+    evidence.
+
+    **When to file**:
+
+    - A dependency is pinned to a version covered by a published CVE.
+    - You have verified the CVE ID and the installed version falls in the
+      affected range (use ``web_search`` if unsure).
+
+    **When NOT to file**:
+
+    - Dynamically-proven vulnerabilities → use
+      ``create_vulnerability_report`` (``finding_class`` dynamic).
+    - Outdated-but-not-vulnerable dependencies with no CVE.
+    - Re-reporting the same CVE/package already filed.
+
+    **Reachability**: do NOT silently downgrade or suppress a finding
+    because the vulnerable code path may be unreachable — instead state
+    reachability as an ``assumptions`` / confidence factor. Report the
+    finding; let the reader weigh exploitability.
+
+    **Formatting**: use markdown in text fields (``**bold**``, ``inline
+    code`` for package/version identifiers, fenced code blocks for
+    manifest excerpts). No internal paths/tooling/agent references.
+
+    Args:
+        title: e.g. ``"CVE-2024-1234 in lodash 4.17.20 (prototype pollution)"``.
+        description: What the CVE is and why the pinned version is affected.
+        target: Affected repository / project / manifest.
+        cve: ``CVE-YYYY-NNNNN`` — required and must be verified.
+        package_name: Affected package name (e.g. ``lodash``).
+        installed_version: The version currently pinned/installed.
+        impact: What the CVE enables; business risk in this context.
+        remediation_steps: How to fix (usually upgrade to a fixed version).
+        assumptions: Exploitability/reachability assumptions & confidence.
+        package_ecosystem: e.g. ``npm`` / ``pypi`` / ``maven`` / ``go``.
+        fixed_version: First non-vulnerable version, if known.
+        cwe: ``CWE-NNN`` (most specific) if certain, else omit.
+        advisory_cvss: **Required.** Published advisory base score
+            (0.0-10.0) — read it off the advisory (trivy CVSS / NVD / GHSA).
+            Severity is derived solely from this score, so it must be the
+            real published value; do not guess or omit it.
+        technical_analysis: Optional deeper mechanism/root-cause detail.
+        fix_effort: One of ``trivial`` / ``low`` / ``medium`` / ``high``
+            (dependency upgrades are usually ``trivial``/``low``).
+    """
+    inner = ctx.context if isinstance(ctx.context, dict) else {}
+    raw_agent_id = inner.get("agent_id")
+    agent_id = raw_agent_id if isinstance(raw_agent_id, str) else None
+    agent_name = None
+    coordinator = inner.get("coordinator")
+    if agent_id is not None and coordinator is not None:
+        names = getattr(coordinator, "names", {})
+        if isinstance(names, dict):
+            raw_agent_name = names.get(agent_id)
+            agent_name = raw_agent_name if isinstance(raw_agent_name, str) else None
+
+    result = await _do_create_dependency(
+        title=title,
+        description=description,
+        target=target,
+        cve=cve,
+        package_name=package_name,
+        installed_version=installed_version,
+        impact=impact,
+        remediation_steps=remediation_steps,
+        assumptions=assumptions,
+        package_ecosystem=package_ecosystem,
+        fixed_version=fixed_version,
+        cwe=cwe,
+        advisory_cvss=advisory_cvss,
+        technical_analysis=technical_analysis,
+        fix_effort=fix_effort,
        agent_id=agent_id,
        agent_name=agent_name,
    )
@@ -5,6 +5,23 @@ invocation the agent makes (nmap, ffuf, agent-browser, python3, …) goes
 through `exec_command`. `write_stdin` streams input to a still-running
 process started by an earlier `exec_command` (for interactive prompts).

+## `write_stdin` requires a TTY-backed process
+
+`exec_command` runs each command in a fresh **non-interactive** shell (plain
+pipes, no TTY) by default. `write_stdin` only works against a process that is
+still running **and** was started with a PTY. The canonical sequence is:
+
+```text
+exec_command(cmd="python3", tty=true)      # start a PTY-backed process
+write_stdin(session_id=<id>, chars="print(1)\n")
+```
+
+Calling `write_stdin` on a command started with the default `tty=false`, or on
+a process that has already exited, fails with
+`stdin is not available for this process. Start the command with 'tty=true' in
+'exec_command' before using 'write_stdin'.` Use `tty=true` for REPLs,
+`ssh`/`nc`/`ftp`, `msfconsole`, or to deliver a Ctrl-C to a long-running job.
+
 - **Implementation:** `agents.sandbox.capabilities.tools.shell_tool.ShellTool`
  (in the upstream `agents` SDK)
 - **Wired in:** `strix/agents/factory.py` — added per-run via the SDK
@@ -0,0 +1,12 @@
+"""Local web viewer for Strix runs.
+
+Serves a prebuilt single-page app that renders a run (live or finished) read
+directly from the run's on-disk files. No cloud dependency, no file picker.
+"""
+
+from __future__ import annotations
+
+from strix.viewer.server import serve
+
+
+__all__ = ["serve"]
@@ -0,0 +1,251 @@
+"""Viewer email verification state and the relay client.
+
+The local viewer proxies email verification and encrypted-report delivery to
+the Strix relay (``STRIX_APP_URL``). The browser never talks to the relay
+directly, and the report password generated locally is never sent to it.
+
+State lives in ``~/.strix/viewer-auth.json`` (0600). ``is_verified`` is a local
+flag that unlocks browsing the run history list; the relay still enforces token
+expiry when a report is actually sent.
+"""
+
+from __future__ import annotations
+
+import base64
+import contextlib
+import json
+import logging
+import urllib.error
+import urllib.request
+from datetime import UTC, datetime
+from pathlib import Path
+from typing import Any
+
+from strix.config.loader import load_settings
+
+
+logger = logging.getLogger(__name__)
+
+AUTH_PATH = Path.home() / ".strix" / "viewer-auth.json"
+
+_OTP_TIMEOUT = 15
+_SEND_TIMEOUT = 30
+
+
+class RelayError(Exception):
+    """A relay call failed. ``code`` is a stable, machine-readable reason."""
+
+    def __init__(self, code: str, message: str | None = None) -> None:
+        self.code = code
+        super().__init__(message or code)
+
+
+# --- local state ------------------------------------------------------------
+
+
+def read_auth() -> dict[str, Any] | None:
+    """Return the stored ``{email, token, verified_at}`` record, or None."""
+    try:
+        data = json.loads(AUTH_PATH.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError):
+        return None
+    if not isinstance(data, dict):
+        return None
+    email = data.get("email")
+    token = data.get("token")
+    if not isinstance(email, str) or not email or not isinstance(token, str) or not token:
+        return None
+    return data
+
+
+def parse_expiry(raw: object) -> datetime | None:
+    """Parse a relay ``expires_at`` value into an aware UTC datetime.
+
+    Accepts both ISO 8601 strings and epoch seconds (as a number or numeric
+    string) so a valid relay expiry is not misread as missing. Returns None only
+    when it is genuinely absent or unparseable; both the local gate (see
+    ``is_verified``) and OTP verification (see ``otp_verify``) fail closed on such
+    values, matching the relay, which rejects a token with no valid expiry.
+    """
+    if isinstance(raw, bool):
+        return None
+    if isinstance(raw, int | float):
+        return _from_epoch(raw)
+    if not isinstance(raw, str) or not raw:
+        return None
+    try:
+        return _from_epoch(float(raw))
+    except ValueError:
+        pass
+    try:
+        parsed = datetime.fromisoformat(raw.replace("Z", "+00:00"))
+    except ValueError:
+        return None
+    return parsed if parsed.tzinfo else parsed.replace(tzinfo=UTC)
+
+
+def _expiry(record: dict[str, Any]) -> datetime | None:
+    """The stored ``verified_at`` parsed to a datetime, or None if unusable."""
+    return parse_expiry(record.get("verified_at"))
+
+
+def _from_epoch(seconds: float) -> datetime | None:
+    """Epoch seconds → aware UTC datetime, or None if out of range."""
+    try:
+        return datetime.fromtimestamp(seconds, tz=UTC)
+    except (OverflowError, OSError, ValueError):
+        return None
+
+
+def is_verified() -> bool:
+    """True when a usable email + token record with a valid future expiry exists.
+
+    The expiry returned by OTP verification is enforced here so history stops
+    unlocking once the token lapses. It fails closed: a record whose expiry is
+    absent, blank, or unparseable requires re-verification rather than unlocking
+    forever, keeping the local gate in step with the relay (which rejects an
+    expired token on report send).
+    """
+    record = read_auth()
+    if record is None:
+        return False
+    expiry = _expiry(record)
+    return expiry is not None and expiry > datetime.now(UTC)
+
+
+def write_auth(email: str, token: str, verified_at: str) -> None:
+    """Atomically persist the auth record with 0600 permissions."""
+    AUTH_PATH.parent.mkdir(parents=True, exist_ok=True)
+    payload = json.dumps({"email": email, "token": token, "verified_at": verified_at})
+    tmp = AUTH_PATH.with_suffix(".json.tmp")
+    tmp.write_text(payload, encoding="utf-8")
+    with contextlib.suppress(OSError):
+        tmp.chmod(0o600)
+    tmp.replace(AUTH_PATH)
+    with contextlib.suppress(OSError):
+        AUTH_PATH.chmod(0o600)
+
+
+def forget() -> None:
+    """Delete the stored auth record. No-op if it is absent."""
+    with contextlib.suppress(OSError):
+        AUTH_PATH.unlink()
+
+
+# --- relay client -----------------------------------------------------------
+
+
+def _app_url() -> str:
+    return load_settings().viewer.app_url.rstrip("/")
+
+
+def _post_json(path: str, payload: dict[str, Any], *, timeout: int) -> tuple[int, dict[str, Any]]:
+    """POST JSON to the relay. Returns (status, parsed body).
+
+    Raises RelayError("unavailable") for network/transport failures. HTTP
+    error responses (4xx/5xx) are returned as (status, body) for the caller to
+    map, not raised.
+    """
+    url = f"{_app_url()}{path}"
+    body = json.dumps(payload).encode("utf-8")
+    request = urllib.request.Request(  # noqa: S310 - fixed https relay URL
+        url,
+        data=body,
+        headers={"Content-Type": "application/json", "Accept": "application/json"},
+        method="POST",
+    )
+    try:
+        with urllib.request.urlopen(request, timeout=timeout) as response:  # noqa: S310
+            return response.status, _parse_body(response.read())
+    except urllib.error.HTTPError as exc:
+        return exc.code, _parse_body(exc.read())
+    except (urllib.error.URLError, TimeoutError, OSError) as exc:
+        logger.warning("relay request to %s failed: %s", path, exc)
+        raise RelayError("unavailable") from exc
+
+
+def _parse_body(raw: bytes) -> dict[str, Any]:
+    try:
+        data = json.loads(raw or b"{}")
+    except json.JSONDecodeError:
+        return {}
+    return data if isinstance(data, dict) else {}
+
+
+def otp_start(email: str) -> None:
+    """Ask the relay to email a verification code. Raises RelayError on failure."""
+    status, data = _post_json("/api/oss/otp/start", {"email": email}, timeout=_OTP_TIMEOUT)
+    if status == 200:
+        return
+    if status == 429:
+        raise RelayError("rate_limited")
+    if status == 400:
+        # The relay uses 400 both for a malformed address and, separately, to
+        # reject a free/personal email domain (it wants a work email).
+        if data.get("error") == "work_email_required":
+            raise RelayError("work_email_required")
+        raise RelayError("invalid_email")
+    raise RelayError("unavailable")
+
+
+def otp_verify(email: str, code: str) -> dict[str, Any]:
+    """Verify a code. Returns ``{token, email, expires_at}`` or raises RelayError."""
+    status, data = _post_json(
+        "/api/oss/otp/verify",
+        {"email": email, "code": code},
+        timeout=_OTP_TIMEOUT,
+    )
+    if status == 200 and isinstance(data.get("token"), str):
+        # A token with no usable expiry cannot unlock history locally (the gate
+        # fails closed), so treat such a response as a failed verification rather
+        # than reporting success and then leaving the user stuck unverified.
+        if parse_expiry(data.get("expires_at")) is None:
+            raise RelayError("unavailable")
+        return data
+    if status == 403:
+        raise RelayError("invalid_code")
+    raise RelayError("unavailable")
+
+
+def report_send(
+    token: str,
+    pdf_bytes: bytes,
+    filename: str,
+    run_name: str,
+    target: str,
+) -> None:
+    """Forward the encrypted PDF to the relay for delivery.
+
+    The report password is NEVER part of this payload; only the encrypted PDF
+    bytes travel to the relay.
+    """
+    payload = {
+        "token": token,
+        "pdf_base64": base64.b64encode(pdf_bytes).decode("ascii"),
+        "filename": filename,
+        "run_name": run_name,
+        "target": target,
+    }
+    status, _ = _post_json("/api/oss/report/send", payload, timeout=_SEND_TIMEOUT)
+    if status == 200:
+        return
+    if status == 401:
+        raise RelayError("reverify")
+    if status == 413:
+        raise RelayError("too_large")
+    if status == 403:
+        raise RelayError("forbidden")
+    raise RelayError("unavailable")
+
+
+__all__ = [
+    "AUTH_PATH",
+    "RelayError",
+    "forget",
+    "is_verified",
+    "otp_start",
+    "otp_verify",
+    "read_auth",
+    "report_send",
+    "write_auth",
+]
@@ -0,0 +1,139 @@
+"""`strix view [<run>]` command: serve a run's viewer UI locally."""
+
+from __future__ import annotations
+
+import argparse
+import logging
+import time
+from typing import TYPE_CHECKING
+
+from rich.console import Console
+
+from strix.core.paths import (
+    RUNS_DIR_NAME,
+    latest_run_dir,
+    run_dir_for,
+    run_record_path,
+    runs_base_dir,
+)
+from strix.viewer.server import authorized_url, bundle_is_built, serve
+from strix.viewer.transcript import read_run_summary
+
+
+if TYPE_CHECKING:
+    from pathlib import Path
+    from typing import NoReturn
+
+
+logger = logging.getLogger(__name__)
+
+
+def run_view(argv: list[str]) -> None:
+    parser = argparse.ArgumentParser(
+        prog="strix view",
+        description="Open a local web view of a Strix run (live or finished).",
+    )
+    parser.add_argument(
+        "run",
+        nargs="?",
+        default=None,
+        help=f"Run name under ./{RUNS_DIR_NAME} (defaults to the most recent run).",
+    )
+    parser.add_argument(
+        "--port",
+        type=int,
+        default=0,
+        help="Port to serve on (default: an available ephemeral port).",
+    )
+    parser.add_argument("--host", default="127.0.0.1", help=argparse.SUPPRESS)
+    parser.add_argument(
+        "--no-open",
+        action="store_true",
+        help="Do not open the browser automatically.",
+    )
+    args = parser.parse_args(argv)
+
+    console = Console()
+
+    if not bundle_is_built():
+        console.print(
+            "[bold red]Viewer UI is not built.[/]\n"
+            "Build it with: [cyan]cd strix/viewer_src && npm ci && npm run build[/]"
+        )
+        raise SystemExit(1)
+
+    run_dir = _resolve_run_dir(args.run, console)
+
+    httpd, url, token = serve(
+        run_dir,
+        host=args.host,
+        port=args.port,
+        open_browser=not args.no_open,
+    )
+    # The tokened URL is what authorizes the browser (steering, report sending,
+    # history). Print it rather than the bare URL so the operator -- and only
+    # the operator -- can open or share an authorized link.
+    open_url = authorized_url(url, token)
+
+    run_name = run_dir.name
+    summary = read_run_summary(run_dir)
+    live = not summary.get("finished", False)
+
+    from strix.telemetry import posthog
+
+    posthog.viewer_opened(source="cli", live=live)
+
+    state_label = "[#eab308]live[/]" if live else "[#22c55e]finished[/]"
+    console.print()
+    console.print(f"Serving [bold white]{run_name}[/] ({state_label}) at [#60a5fa]{open_url}[/]")
+    console.print("[dim]This link authorizes the browser; anyone you share it with can steer[/]")
+    console.print("[dim]a live scan and browse history. Press Ctrl-C to stop the viewer.[/]")
+    console.print()
+
+    try:
+        while True:
+            time.sleep(1.0)
+    except KeyboardInterrupt:
+        console.print("\n[dim]Viewer stopped.[/]")
+    finally:
+        httpd.shutdown()
+        httpd.server_close()
+
+
+def _resolve_run_dir(run: str | None, console: Console) -> Path:
+    if run:
+        run_dir = run_dir_for(run)
+        if not run_record_path(run_dir).is_file():
+            _fail_no_run(console, requested=run)
+        return run_dir
+
+    latest = latest_run_dir()
+    if latest is None:
+        _fail_no_run(console, requested=None)
+    return latest
+
+
+def _fail_no_run(console: Console, *, requested: str | None) -> NoReturn:
+    base = runs_base_dir()
+    available = (
+        sorted(
+            (child.name for child in base.iterdir() if run_record_path(child).is_file()),
+            reverse=True,
+        )
+        if base.is_dir()
+        else []
+    )
+
+    if requested:
+        console.print(f"[bold red]No run named '{requested}' under ./{RUNS_DIR_NAME}.[/]")
+    else:
+        console.print(f"[bold red]No runs found under ./{RUNS_DIR_NAME}.[/]")
+
+    if available:
+        console.print("Available runs:")
+        for name in available[:20]:
+            console.print(f"  [cyan]{name}[/]")
+    raise SystemExit(1)
+
+
+__all__ = ["run_view"]
@@ -0,0 +1,618 @@
+"""Build and encrypt a branded PDF report for a run.
+
+The layout mirrors the Strix cloud pentest report (cover page, executive
+severity grid, per-finding detail with colored severity badges) but is rendered
+entirely locally with reportlab, so it ships without a browser or heavy system
+deps and keeps the report on the user's machine.
+
+The PDF carries FULL finding detail, including proof-of-concept scripts, so it
+is encrypted end to end with AES-256. The password is generated locally with a
+CSPRNG, shown only to the local browser, and never leaves the machine except in
+the user's own hands. Strix cannot read the delivered report.
+"""
+
+from __future__ import annotations
+
+import html
+import re
+import secrets
+from datetime import datetime
+from io import BytesIO
+from typing import TYPE_CHECKING, Any
+
+from pypdf import PdfReader, PdfWriter
+from reportlab.lib import colors
+from reportlab.lib.enums import TA_CENTER
+from reportlab.lib.pagesizes import A4
+from reportlab.lib.styles import ParagraphStyle
+from reportlab.lib.units import mm
+from reportlab.pdfgen import canvas as pdfcanvas
+from reportlab.platypus import (
+    Flowable,
+    KeepTogether,
+    PageBreak,
+    Paragraph,
+    SimpleDocTemplate,
+    Spacer,
+    Table,
+    TableStyle,
+)
+
+from strix.viewer.transcript import (
+    primary_target,
+    read_run_summary,
+    read_vulnerabilities,
+    severity_counts,
+)
+
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+
+# Palette lifted from the cloud report theme (styles/base.ts, docx/theme.ts).
+_INK = colors.HexColor("#000000")
+_TEXT = colors.HexColor("#1a1a1a")
+_MUTED = colors.HexColor("#666666")
+_FAINT = colors.HexColor("#999999")
+_BORDER = colors.HexColor("#e5e5e5")
+_LIGHT_BG = colors.HexColor("#f7f7f7")
+
+_SEVERITY_ORDER = ("critical", "high", "medium", "low")
+_SEVERITY_COLORS = {
+    "critical": colors.HexColor("#dc2626"),
+    "high": colors.HexColor("#ea580c"),
+    "medium": colors.HexColor("#ca8a04"),
+    "low": colors.HexColor("#2563eb"),
+}
+
+# Helvetica stands in for Geist: a clean sans with no font file to ship.
+_SANS = "Helvetica"
+_SANS_BOLD = "Helvetica-Bold"
+_MONO = "Courier"
+
+_PAGE_W, _PAGE_H = A4
+
+
+def _esc(value: Any) -> str:
+    """Escape a value for reportlab's Paragraph markup."""
+    return html.escape(str(value)).replace("\n", "<br/>")
+
+
+class _NumberedCanvas(pdfcanvas.Canvas):  # type: ignore[misc]  # reportlab base is untyped
+    """Two-pass canvas that prints 'Page X of Y' on every page after the cover."""
+
+    def __init__(self, *args: Any, **kwargs: Any) -> None:
+        super().__init__(*args, **kwargs)
+        self._saved_states: list[dict[str, Any]] = []
+
+    def showPage(self) -> None:  # noqa: N802 - reportlab API
+        self._saved_states.append(dict(self.__dict__))
+        self._startPage()
+
+    def save(self) -> None:
+        total = len(self._saved_states)
+        for index, state in enumerate(self._saved_states):
+            self.__dict__.update(state)
+            if index > 0:  # skip the cover page
+                self._draw_footer(index + 1, total)
+            super().showPage()
+        super().save()
+
+    def _draw_footer(self, page: int, total: int) -> None:
+        self.setFont(_SANS, 8)
+        self.setFillColor(_FAINT)
+        self.drawCentredString(_PAGE_W / 2, 14 * mm, f"Page {page} of {total}")
+
+
+class _LogoMark(Flowable):  # type: ignore[misc]  # reportlab base is untyped
+    """The rounded-square Strix mark drawn inline (no raster asset to ship)."""
+
+    def __init__(self, size: float = 30) -> None:
+        super().__init__()
+        self.size = size
+        self.width = size
+        self.height = size
+
+    def draw(self) -> None:
+        c = self.canv
+        s = self.size
+        c.setFillColor(_INK)
+        c.roundRect(0, 0, s, s, s * 0.28, fill=1, stroke=0)
+        c.setFillColor(colors.white)
+        c.setFont(_SANS_BOLD, s * 0.56)
+        c.drawCentredString(s / 2, s * 0.27, "S")
+
+
+def _styles() -> dict[str, ParagraphStyle]:
+    styles: dict[str, ParagraphStyle] = {}
+    styles["wordmark"] = ParagraphStyle(
+        "Wordmark", fontName=_SANS_BOLD, fontSize=17, leading=20, textColor=_INK
+    )
+    styles["badge_label"] = ParagraphStyle(
+        "BadgeLabel", fontName=_SANS_BOLD, fontSize=9, leading=12, textColor=_MUTED
+    )
+    styles["cover_title"] = ParagraphStyle(
+        "CoverTitle", fontName=_SANS_BOLD, fontSize=34, leading=38, textColor=_INK
+    )
+    styles["cover_org"] = ParagraphStyle(
+        "CoverOrg", fontName=_SANS, fontSize=13, leading=18, textColor=_MUTED
+    )
+    styles["meta_label"] = ParagraphStyle(
+        "MetaLabel", fontName=_SANS_BOLD, fontSize=8, leading=12, textColor=_MUTED
+    )
+    styles["meta_value"] = ParagraphStyle(
+        "MetaValue", fontName=_SANS, fontSize=10.5, leading=14, textColor=_TEXT
+    )
+    styles["section"] = ParagraphStyle(
+        "Section", fontName=_SANS_BOLD, fontSize=18, leading=22, textColor=_INK, spaceAfter=6
+    )
+    styles["finding"] = ParagraphStyle(
+        "Finding", fontName=_SANS_BOLD, fontSize=13, leading=17, textColor=_INK, spaceBefore=6
+    )
+    styles["field_label"] = ParagraphStyle(
+        "FieldLabel", fontName=_SANS_BOLD, fontSize=8.5, leading=12, textColor=_MUTED,
+        spaceBefore=10, spaceAfter=2,
+    )
+    styles["body"] = ParagraphStyle(
+        "Body", fontName=_SANS, fontSize=10, leading=15, textColor=_TEXT, spaceAfter=8
+    )
+    styles["md_heading"] = ParagraphStyle(
+        "MdHeading", fontName=_SANS_BOLD, fontSize=11, leading=15, textColor=_INK,
+        spaceBefore=10, spaceAfter=4,
+    )
+    styles["bullet"] = ParagraphStyle(
+        "Bullet", fontName=_SANS, fontSize=10, leading=15, textColor=_TEXT,
+        leftIndent=16, firstLineIndent=-11, spaceAfter=3,
+    )
+    styles["meta_inline"] = ParagraphStyle(
+        "MetaInline", fontName=_SANS, fontSize=9, leading=13, textColor=_MUTED, spaceBefore=4
+    )
+    # spaceBefore/spaceAfter must exceed borderPadding: reportlab does not reserve
+    # a bordered paragraph's top padding, so too small a gap lets the background
+    # box bleed up over the field label above it.
+    styles["code"] = ParagraphStyle(
+        "Code", fontName=_MONO, fontSize=8, leading=11, textColor=_TEXT,
+        backColor=_LIGHT_BG, borderColor=_BORDER, borderWidth=0.5, borderPadding=8,
+        spaceBefore=12, spaceAfter=12,
+    )
+    styles["count"] = ParagraphStyle(
+        "Count", fontName=_SANS_BOLD, fontSize=30, leading=32, alignment=TA_CENTER
+    )
+    styles["count_label"] = ParagraphStyle(
+        "CountLabel", fontName=_SANS_BOLD, fontSize=8, leading=12, textColor=_MUTED,
+        alignment=TA_CENTER, spaceBefore=4,
+    )
+    styles["badge"] = ParagraphStyle(
+        "Badge", fontName=_SANS_BOLD, fontSize=9, leading=11, textColor=colors.white,
+        alignment=TA_CENTER,
+    )
+    styles["confidential"] = ParagraphStyle(
+        "Confidential", fontName=_SANS_BOLD, fontSize=9, leading=12, textColor=colors.white,
+        alignment=TA_CENTER,
+    )
+    return styles
+
+
+def _parse_time(raw: Any) -> datetime | None:
+    if not isinstance(raw, str) or not raw:
+        return None
+    text = raw.strip().replace(" UTC", "Z").replace(" ", "T")
+    if text.endswith("Z"):
+        text = text[:-1] + "+00:00"
+    try:
+        return datetime.fromisoformat(text)
+    except ValueError:
+        return None
+
+
+def _fmt_time(raw: Any) -> str:
+    parsed = _parse_time(raw)
+    return parsed.strftime("%Y-%m-%d %H:%M UTC") if parsed else "n/a"
+
+
+def _duration(start: Any, end: Any) -> str:
+    start_dt = _parse_time(start)
+    end_dt = _parse_time(end)
+    if not start_dt or not end_dt:
+        return "n/a"
+    seconds = int((end_dt - start_dt).total_seconds())
+    if seconds < 0:
+        return "n/a"
+    hours, remainder = divmod(seconds, 3600)
+    minutes, secs = divmod(remainder, 60)
+    if hours:
+        return f"{hours}h {minutes}m {secs}s"
+    if minutes:
+        return f"{minutes}m {secs}s"
+    return f"{secs}s"
+
+
+def _severity_badge(styles: dict[str, ParagraphStyle], severity: str) -> Table:
+    """A colored pill matching .severity-badge in the cloud report."""
+    color = _SEVERITY_COLORS.get(severity, _MUTED)
+    cell = Paragraph(severity.upper(), styles["badge"])
+    table = Table([[cell]], colWidths=[len(severity) * 6.5 + 20])
+    table.setStyle(
+        TableStyle(
+            [
+                ("BACKGROUND", (0, 0), (-1, -1), color),
+                ("TOPPADDING", (0, 0), (-1, -1), 4),
+                ("BOTTOMPADDING", (0, 0), (-1, -1), 4),
+                ("LEFTPADDING", (0, 0), (-1, -1), 8),
+                ("RIGHTPADDING", (0, 0), (-1, -1), 8),
+                ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
+            ]
+        )
+    )
+    table.hAlign = "LEFT"
+    return table
+
+
+def _severity_grid(styles: dict[str, ParagraphStyle], counts: dict[str, int]) -> Table:
+    """The four-card severity grid from the executive summary."""
+    cells: list[list[Flowable]] = []
+    for name in _SEVERITY_ORDER:
+        color = _SEVERITY_COLORS[name]
+        count_style = ParagraphStyle(f"Count{name}", parent=styles["count"], textColor=color)
+        cells.append(
+            [Paragraph(str(counts.get(name, 0)), count_style),
+             Paragraph(name.upper(), styles["count_label"])]
+        )
+    col = (_PAGE_W - 40 * mm) / 4
+    table = Table([cells], colWidths=[col] * 4)
+    style = [
+        ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
+        ("TOPPADDING", (0, 0), (-1, -1), 16),
+        ("BOTTOMPADDING", (0, 0), (-1, -1), 16),
+        ("GRID", (0, 0), (-1, -1), 0.5, _BORDER),
+    ]
+    for index, name in enumerate(_SEVERITY_ORDER):
+        style.append(("LINEABOVE", (index, 0), (index, 0), 3, _SEVERITY_COLORS[name]))
+    table.setStyle(TableStyle(style))
+    return table
+
+
+def _section(styles: dict[str, ParagraphStyle], title: str) -> Table:
+    """Section title with the underline rule from h2.section-title."""
+    table = Table([[Paragraph(_esc(title), styles["section"])]], colWidths=[_PAGE_W - 40 * mm])
+    table.setStyle(
+        TableStyle(
+            [
+                ("LINEBELOW", (0, 0), (-1, -1), 1, _BORDER),
+                ("BOTTOMPADDING", (0, 0), (-1, -1), 10),
+                ("LEFTPADDING", (0, 0), (-1, -1), 0),
+                ("RIGHTPADDING", (0, 0), (-1, -1), 0),
+                ("TOPPADDING", (0, 0), (-1, -1), 0),
+            ]
+        )
+    )
+    return table
+
+
+def _cover(
+    styles: dict[str, ParagraphStyle], record: dict[str, Any], run_name: str
+) -> list[Flowable]:
+    header = Table(
+        [[_LogoMark(30), Paragraph("Strix", styles["wordmark"])]],
+        colWidths=[38, _PAGE_W - 40 * mm - 38],
+    )
+    header.setStyle(
+        TableStyle(
+            [
+                ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
+                ("LEFTPADDING", (0, 0), (-1, -1), 0),
+                ("RIGHTPADDING", (0, 0), (-1, -1), 0),
+                ("TOPPADDING", (0, 0), (-1, -1), 0),
+                ("BOTTOMPADDING", (0, 0), (-1, -1), 0),
+            ]
+        )
+    )
+
+    target = primary_target(record) or "Target"
+    meta_rows = [
+        ("TARGET", primary_target(record) or "unknown target"),
+        ("RUN", run_name),
+        ("SCAN MODE", str(record.get("scan_mode") or "n/a")),
+        ("STATUS", str(record.get("status") or "n/a")),
+        ("STARTED", _fmt_time(record.get("start_time"))),
+        ("COMPLETED", _fmt_time(record.get("end_time"))),
+        ("DURATION", _duration(record.get("start_time"), record.get("end_time"))),
+    ]
+    meta_table = Table(
+        [[Paragraph(label, styles["meta_label"]), Paragraph(_esc(value), styles["meta_value"])]
+         for label, value in meta_rows],
+        colWidths=[38 * mm, _PAGE_W - 40 * mm - 38 * mm],
+    )
+    meta_table.setStyle(
+        TableStyle(
+            [
+                ("VALIGN", (0, 0), (-1, -1), "TOP"),
+                ("LEFTPADDING", (0, 0), (-1, -1), 0),
+                ("TOPPADDING", (0, 0), (-1, -1), 6),
+                ("BOTTOMPADDING", (0, 0), (-1, -1), 6),
+                ("LINEBELOW", (0, 0), (-1, -2), 0.5, _BORDER),
+            ]
+        )
+    )
+
+    confidential = Table([[Paragraph("CONFIDENTIAL", styles["confidential"])]], colWidths=[120])
+    confidential.setStyle(
+        TableStyle(
+            [
+                ("BACKGROUND", (0, 0), (-1, -1), _INK),
+                ("TOPPADDING", (0, 0), (-1, -1), 8),
+                ("BOTTOMPADDING", (0, 0), (-1, -1), 8),
+                ("VALIGN", (0, 0), (-1, -1), "MIDDLE"),
+            ]
+        )
+    )
+    confidential.hAlign = "CENTER"
+
+    return [
+        header,
+        Spacer(1, 150),
+        Paragraph("PENETRATION TEST REPORT", styles["badge_label"]),
+        Spacer(1, 20),
+        Paragraph("Security Assessment", styles["cover_title"]),
+        Paragraph(_esc(target), styles["cover_org"]),
+        Spacer(1, 28),
+        meta_table,
+        Spacer(1, 90),
+        confidential,
+        PageBreak(),
+    ]
+
+
+def _inline_md(text: str) -> str:
+    """Convert inline markdown (bold, italic, `code`) to reportlab markup.
+
+    Code spans are stashed as placeholders before bold/italic run, so bold that
+    wraps a code span (``**`x`**``) works and code contents are never mangled.
+    """
+    codes: list[str] = []
+
+    def _stash(match: re.Match[str]) -> str:
+        codes.append(match.group(1))
+        return f"\x00{len(codes) - 1}\x00"
+
+    seg = html.escape(re.sub(r"`([^`]+)`", _stash, text))
+    seg = re.sub(r"\*\*(.+?)\*\*", r"<b>\1</b>", seg)
+    seg = re.sub(r"__(.+?)__", r"<b>\1</b>", seg)
+    seg = re.sub(r"\*(.+?)\*", r"<i>\1</i>", seg)
+
+    def _restore(match: re.Match[str]) -> str:
+        inner = html.escape(codes[int(match.group(1))])
+        return f'<font face="{_MONO}" color="#b31d28">{inner}</font>'
+
+    return re.sub(r"\x00(\d+)\x00", _restore, seg)
+
+
+def _strip_leading_heading(md: str) -> str:
+    """Drop a single leading markdown heading (each section adds its own title)."""
+    lines = md.lstrip("\n").split("\n")
+    if lines and re.match(r"^#{1,6}\s+", lines[0].strip()):
+        return "\n".join(lines[1:]).lstrip("\n")
+    return md
+
+
+def _markdown_flowables(  # noqa: PLR0915 - cohesive block parser, splitting hurts clarity
+    md: str, styles: dict[str, ParagraphStyle]
+) -> list[Flowable]:
+    """Render a markdown block (headings, lists, fenced code, prose) to flowables."""
+    flow: list[Flowable] = []
+    para: list[str] = []
+    bullets: list[tuple[str, str]] = []
+
+    def flush_para() -> None:
+        if para:
+            flow.append(Paragraph(_inline_md(" ".join(para)), styles["body"]))
+            para.clear()
+
+    def flush_bullets() -> None:
+        for marker, item in bullets:
+            flow.append(Paragraph(f"{marker}&nbsp;{_inline_md(item)}", styles["bullet"]))
+        bullets.clear()
+
+    lines = md.replace("\r\n", "\n").split("\n")
+    i = 0
+    while i < len(lines):
+        stripped = lines[i].strip()
+        if stripped.startswith("```"):
+            flush_para()
+            flush_bullets()
+            i += 1
+            code: list[str] = []
+            while i < len(lines) and not lines[i].strip().startswith("```"):
+                code.append(lines[i])
+                i += 1
+            i += 1  # closing fence
+            flow.append(Paragraph(_esc("\n".join(code)) or "&nbsp;", styles["code"]))
+            continue
+        if not stripped:
+            flush_para()
+            flush_bullets()
+            i += 1
+            continue
+        heading = re.match(r"^(#{1,6})\s+(.*)$", stripped)
+        if heading:
+            flush_para()
+            flush_bullets()
+            flow.append(Paragraph(_inline_md(heading.group(2)), styles["md_heading"]))
+            i += 1
+            continue
+        ordered = re.match(r"^(\d+)\.\s+(.*)$", stripped)
+        unordered = re.match(r"^[-*+]\s+(.*)$", stripped)
+        if ordered:
+            flush_para()
+            bullets.append((f"{ordered.group(1)}.", ordered.group(2)))
+            i += 1
+            continue
+        if unordered:
+            flush_para()
+            bullets.append(("•", unordered.group(1)))
+            i += 1
+            continue
+        flush_bullets()
+        para.append(stripped)
+        i += 1
+
+    flush_para()
+    flush_bullets()
+    return flow
+
+
+def _field_block(
+    styles: dict[str, ParagraphStyle], label: str, value: Any, *, code: bool = False
+) -> list[Flowable]:
+    if value is None or (isinstance(value, str) and not value.strip()):
+        return []
+    flow: list[Flowable] = [Paragraph(label.upper(), styles["field_label"])]
+    if code:
+        flow.append(Paragraph(_esc(value), styles["code"]))
+    else:
+        flow.extend(_markdown_flowables(str(value), styles))
+    return flow
+
+
+def _finding_flowables(
+    styles: dict[str, ParagraphStyle], index: int, vuln: dict[str, Any]
+) -> list[Flowable]:
+    title = vuln.get("title") or "Untitled finding"
+    severity = str(vuln.get("severity") or "").lower().strip() or "low"
+
+    meta_bits = []
+    if vuln.get("cvss") is not None:
+        meta_bits.append(f"<b>CVSS</b> {_esc(vuln.get('cvss'))}")
+    meta_bits.extend(
+        f"<b>{key.title()}</b> {_esc(vuln.get(key))}"
+        for key in ("target", "endpoint", "method")
+        if vuln.get(key)
+    )
+
+    header: list[Flowable] = [
+        Paragraph(f"{index}. {_esc(title)}", styles["finding"]),
+        Spacer(1, 4),
+        _severity_badge(styles, severity),
+    ]
+    if meta_bits:
+        header.append(Paragraph("&nbsp;&nbsp;".join(meta_bits), styles["meta_inline"]))
+
+    story: list[Flowable] = [KeepTogether(header)]
+    story.extend(_field_block(styles, "Description", vuln.get("description")))
+    story.extend(_field_block(styles, "Impact", vuln.get("impact")))
+    story.extend(_field_block(styles, "Technical analysis", vuln.get("technical_analysis")))
+    story.extend(_field_block(styles, "Proof of concept", vuln.get("poc_description")))
+    story.extend(_field_block(styles, "PoC script", vuln.get("poc_script_code"), code=True))
+    story.extend(_field_block(styles, "Evidence", vuln.get("evidence"), code=True))
+
+    remediation = vuln.get("remediation_steps")
+    if isinstance(remediation, list):
+        remediation = "\n".join(str(step) for step in remediation)
+    story.extend(_field_block(styles, "Remediation", remediation))
+
+    story.append(Spacer(1, 22))
+    return story
+
+
+def _overview_flowables(
+    styles: dict[str, ParagraphStyle], record: dict[str, Any], total: int, counts: dict[str, int]
+) -> list[Flowable]:
+    story: list[Flowable] = [
+        _section(styles, "Executive Summary"),
+        Spacer(1, 16),
+        _severity_grid(styles, counts),
+        Spacer(1, 10),
+        Paragraph(f"<b>{total}</b> total findings across this assessment.", styles["body"]),
+    ]
+    scan_results = record.get("scan_results")
+    if not isinstance(scan_results, dict):
+        return story
+    summary = scan_results.get("executive_summary")
+    if isinstance(summary, str) and summary.strip():
+        story.append(Spacer(1, 16))
+        story.extend(_markdown_flowables(_strip_leading_heading(summary), styles))
+    for label, key in (
+        ("Methodology", "methodology"),
+        ("Technical Analysis", "technical_analysis"),
+        ("Recommendations", "recommendations"),
+    ):
+        value = scan_results.get(key)
+        if isinstance(value, str) and value.strip():
+            story.append(Spacer(1, 20))
+            story.append(_section(styles, label))
+            story.append(Spacer(1, 12))
+            story.extend(_markdown_flowables(_strip_leading_heading(value), styles))
+    return story
+
+
+def generate_report_pdf(run_dir: Path) -> bytes:
+    """Render a branded, full-detail PDF report for the run at ``run_dir``."""
+    record = read_run_summary(run_dir)
+    vulns = [v for v in read_vulnerabilities(run_dir) if isinstance(v, dict)]
+    counts = severity_counts(vulns)
+    run_name = str(record.get("run_name") or run_dir.name)
+
+    styles = _styles()
+    buffer = BytesIO()
+    doc = SimpleDocTemplate(
+        buffer,
+        pagesize=A4,
+        title="Strix Security Report",
+        author="Strix",
+        leftMargin=20 * mm,
+        rightMargin=20 * mm,
+        topMargin=22 * mm,
+        bottomMargin=24 * mm,
+    )
+
+    story: list[Flowable] = []
+    story.extend(_cover(styles, record, run_name))
+    story.extend(_overview_flowables(styles, record, len(vulns), counts))
+
+    story.append(PageBreak())
+    story.append(_section(styles, "Findings"))
+    story.append(Spacer(1, 16))
+    if vulns:
+        for index, vuln in enumerate(vulns, start=1):
+            story.extend(_finding_flowables(styles, index, vuln))
+    else:
+        story.append(Paragraph("No findings were recorded for this run.", styles["body"]))
+
+    doc.build(story, canvasmaker=_NumberedCanvas)
+    return buffer.getvalue()
+
+
+def generate_password() -> str:
+    """Return a >=20 character URL-safe password from a CSPRNG."""
+    return secrets.token_urlsafe(16)
+
+
+def encrypt_pdf(pdf_bytes: bytes, password: str) -> bytes:
+    """Encrypt a PDF with AES-256 using ``password`` as the user password."""
+    reader = PdfReader(BytesIO(pdf_bytes))
+    writer = PdfWriter()
+    writer.append(reader)
+    writer.encrypt(user_password=password, algorithm="AES-256")
+    out = BytesIO()
+    writer.write(out)
+    return out.getvalue()
+
+
+def build_encrypted_report(run_dir: Path) -> tuple[bytes, str, str]:
+    """Build, encrypt, and name the report. Returns (pdf_bytes, password, filename)."""
+    record = read_run_summary(run_dir)
+    run_name = str(record.get("run_name") or run_dir.name)
+    pdf_bytes = generate_report_pdf(run_dir)
+    password = generate_password()
+    encrypted = encrypt_pdf(pdf_bytes, password)
+    filename = f"strix-report-{run_name}.pdf"
+    return encrypted, password, filename
+
+
+__all__ = [
+    "build_encrypted_report",
+    "encrypt_pdf",
+    "generate_password",
+    "generate_report_pdf",
+]
@@ -0,0 +1,537 @@
+"""Local HTTP server that serves the viewer SPA and a run's data from disk.
+
+Design notes:
+- Uses only the standard library (no new runtime dependency). The workload is
+  serving static files plus a handful of JSON reads off disk, so an async stack
+  buys nothing here.
+- The browser polls the JSON endpoints (~1s) rather than using SSE: a finished
+  run stops polling, and short-lived polls survive sleep/network blips without
+  server-side connection state, which suits a stdlib ThreadingHTTPServer.
+- All reads happen per-request straight from disk, so the same server serves a
+  live in-progress run and a finished one identically; the SPA distinguishes
+  them via the ``finished`` flag on /api/run.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import mimetypes
+import secrets
+import threading
+import webbrowser
+from http import HTTPStatus
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from pathlib import Path
+from typing import TYPE_CHECKING, Any
+from urllib.parse import parse_qs, unquote, urlencode, urlsplit
+
+from strix.core.paths import run_record_path
+from strix.viewer import auth
+from strix.viewer.transcript import (
+    build_run_state,
+    primary_target,
+    read_report_markdown,
+    read_run_summary,
+    read_vulnerabilities,
+    severity_counts,
+)
+
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+
+
+logger = logging.getLogger(__name__)
+
+
+def bundle_dir() -> Path:
+    """Directory holding the committed, prebuilt SPA (index.html + assets)."""
+    return Path(__file__).resolve().parent / "viewer_dist"
+
+
+def bundle_is_built() -> bool:
+    return (bundle_dir() / "index.html").is_file()
+
+
+def _iter_run_dirs(base_dir: Path) -> list[Path]:
+    """Every run directory under ``base_dir``, newest first by record mtime."""
+    if not base_dir.is_dir():
+        return []
+    run_dirs = [child for child in base_dir.iterdir() if run_record_path(child).is_file()]
+    run_dirs.sort(key=lambda child: run_record_path(child).stat().st_mtime, reverse=True)
+    return run_dirs
+
+
+def run_list_entry(run_dir: Path) -> dict[str, Any]:
+    """Compact summary of a single run for the history list."""
+    record = read_run_summary(run_dir)
+    return {
+        "name": record.get("run_name") or run_dir.name,
+        "target": primary_target(record),
+        "scan_mode": record.get("scan_mode"),
+        "status": record.get("status"),
+        "start_time": record.get("start_time"),
+        "end_time": record.get("end_time"),
+        "finished": bool(record.get("finished")),
+        "severity_counts": severity_counts(read_vulnerabilities(run_dir)),
+    }
+
+
+def build_runs_payload(base_dir: Path, *, verified: bool) -> dict[str, Any]:
+    """The /api/runs payload. Gates the run list behind email verification.
+
+    The count is always advertised so the UI can tease the history, but the
+    entries only appear once the viewer is verified.
+    """
+    run_dirs = _iter_run_dirs(base_dir)
+    count = len(run_dirs)
+    if not verified:
+        return {"locked": True, "count": count, "runs": []}
+    return {"locked": False, "count": count, "runs": [run_list_entry(d) for d in run_dirs]}
+
+
+def resolve_run_dir(base_dir: Path, run_param: str | None, default_run_dir: Path) -> Path | None:
+    """Resolve a ``?run=`` value to a real run directory under ``base_dir``.
+
+    Returns ``default_run_dir`` when no run is requested. Rejects traversal and
+    unknown runs (returns None) so the caller can answer 404.
+    """
+    if not run_param:
+        return default_run_dir
+    base = base_dir.resolve()
+    candidate = (base / run_param).resolve()
+    # Only direct children of the runs base that actually hold a run record.
+    if candidate.parent != base or not run_record_path(candidate).is_file():
+        return None
+    return candidate
+
+
+# Name of the cookie carrying the per-process session capability.
+SESSION_COOKIE = "strix_viewer_session"
+
+
+class _ViewerState:
+    def __init__(
+        self,
+        run_dir: Path,
+        assets_dir: Path,
+        steer_handler: Callable[[str, str], bool] | None = None,
+    ) -> None:
+        self.run_dir = run_dir
+        self.assets_dir = assets_dir
+        # The strix_runs directory that holds the launched run; used to
+        # enumerate and resolve other runs for the history list.
+        self.base_dir = run_dir.parent
+        # Set only when the viewer runs inside a live scan process (the TUI
+        # launcher), which can deliver a message to a running agent. Absent for
+        # standalone ``strix view`` / finished runs, so steering is unavailable.
+        self.steer_handler = steer_handler
+        # Unguessable per-process capability. It is minted here, printed/opened
+        # for the operator who started the server (see ``authorized_url``), and
+        # exchanged for a session cookie only when presented on the initial page
+        # load. It is the request-level authorization the review asked for:
+        # reachability of the port (e.g. when bound with ``--host``) is not
+        # enough to steer a live scan, trigger a report, or browse history --
+        # the token is never handed to a caller who merely reaches ``/``.
+        self.session_token = secrets.token_urlsafe(32)
+
+
+def _make_handler(state: _ViewerState) -> type[BaseHTTPRequestHandler]:
+    class ViewerHandler(BaseHTTPRequestHandler):
+        server_version = "StrixViewer/1.0"
+
+        def log_message(self, format: str, *args: Any) -> None:  # noqa: A002
+            logger.debug("viewer %s - %s", self.address_string(), format % args)
+
+        def do_GET(self) -> None:
+            parts = urlsplit(self.path)
+            path = parts.path
+            try:
+                if path.startswith("/api/"):
+                    self._handle_api(path, parse_qs(parts.query))
+                else:
+                    self._handle_static(path, parse_qs(parts.query))
+            except BrokenPipeError:
+                # The browser closed the connection mid-response (e.g. it
+                # navigated away between polls). Not an error.
+                logger.debug("viewer client disconnected during %s", path)
+            except Exception:
+                # A bad request must never kill the worker thread.
+                logger.exception("viewer request failed: %s", path)
+                self._send_json(HTTPStatus.INTERNAL_SERVER_ERROR, {"error": "internal error"})
+
+        def do_POST(self) -> None:
+            path = urlsplit(self.path).path
+            try:
+                if path == "/api/event":
+                    self._handle_event()
+                elif path == "/api/auth/otp/start":
+                    self._handle_otp_start()
+                elif path == "/api/auth/otp/verify":
+                    self._handle_otp_verify()
+                elif path == "/api/auth/forget":
+                    self._handle_forget()
+                elif path == "/api/report/send":
+                    self._handle_report_send()
+                elif path == "/api/agents/steer":
+                    self._handle_steer()
+                else:
+                    self._send_json(HTTPStatus.NOT_FOUND, {"error": "unknown endpoint"})
+            except BrokenPipeError:
+                logger.debug("viewer client disconnected during POST %s", path)
+            except Exception:
+                # A bad request must never kill the worker thread.
+                logger.exception("viewer request failed: POST %s", path)
+                self._send_json(HTTPStatus.INTERNAL_SERVER_ERROR, {"error": "internal error"})
+
+        def _read_body(self) -> dict[str, Any]:
+            length = int(self.headers.get("Content-Length") or 0)
+            raw = self.rfile.read(length) if length else b""
+            try:
+                body = json.loads(raw or b"{}")
+            except json.JSONDecodeError:
+                return {}
+            return body if isinstance(body, dict) else {}
+
+        # Funnel events the viewer is allowed to forward. This handler is the
+        # trust boundary: only these event names, with only their known props,
+        # ever reach PostHog. Everything else (including any PII) is dropped.
+        _EMAIL_EVENTS = frozenset(
+            {"email_submitted", "email_verified", "report_sent", "work_email_required"}
+        )
+
+        def _handle_event(self) -> None:
+            body = self._read_body()
+            # Forwarded as anonymous PostHog events that respect the global
+            # telemetry opt-out. Never forward the email, code, or report body:
+            # only the whitelisted event names and their known props are passed.
+            event = body.get("event")
+            if event == "cta_clicked":
+                from strix.telemetry import posthog
+
+                cta = str(body.get("cta") or "unknown")
+                surface = body.get("surface")
+                posthog.viewer_cta_clicked(cta, surface=str(surface) if surface else None)
+            elif event in self._EMAIL_EVENTS:
+                from strix.telemetry import posthog
+
+                purpose = body.get("purpose")
+                posthog.viewer_email_event(str(event), purpose=str(purpose) if purpose else None)
+            self.send_response(HTTPStatus.NO_CONTENT)
+            self.end_headers()
+
+        def _handle_api(self, path: str, query: dict[str, list[str]]) -> None:
+            # The launched run is always viewable with no verification. Only the
+            # cross-run history list (/api/runs) is gated.
+            if path == "/api/runs":
+                payload = build_runs_payload(state.base_dir, verified=auth.is_verified())
+                self._send_json(HTTPStatus.OK, payload)
+                return
+            if path == "/api/capabilities":
+                # Steering is only possible when the viewer shares a live scan's
+                # coordinator + event loop (the TUI launcher wires a handler).
+                self._send_json(HTTPStatus.OK, {"can_steer": state.steer_handler is not None})
+                return
+            if path == "/api/auth/status":
+                # Report verification through is_verified() so an expired record
+                # is advertised as unverified -- otherwise the SPA would suppress
+                # re-verification while history stays locked, stranding the user.
+                record = auth.read_auth()
+                self._send_json(
+                    HTTPStatus.OK,
+                    {
+                        "verified": auth.is_verified(),
+                        "email": record.get("email") if record else None,
+                    },
+                )
+                return
+
+            run_values = query.get("run")
+            run_param = run_values[0] if run_values else None
+            run_dir = resolve_run_dir(state.base_dir, run_param, state.run_dir)
+            if run_dir is None:
+                self._send_json(HTTPStatus.NOT_FOUND, {"error": "unknown run"})
+                return
+
+            # The launched run is always viewable. Any *other* run's data is part
+            # of the gated history: it needs this process's session capability
+            # (so merely reaching an exposed --host port is not enough) *and*
+            # email verification -- otherwise knowing a run name would leak its
+            # metadata, vulnerabilities, report, and transcript.
+            if run_dir.resolve() != state.run_dir.resolve():
+                if not self._has_session():
+                    self._send_json(HTTPStatus.FORBIDDEN, {"error": "forbidden"})
+                    return
+                if not auth.is_verified():
+                    self._send_json(HTTPStatus.UNAUTHORIZED, {"error": "unverified"})
+                    return
+
+            if path == "/api/run":
+                self._send_json(HTTPStatus.OK, read_run_summary(run_dir))
+            elif path == "/api/vulnerabilities":
+                self._send_json(HTTPStatus.OK, read_vulnerabilities(run_dir))
+            elif path == "/api/report":
+                self._send_json(HTTPStatus.OK, {"markdown": read_report_markdown(run_dir)})
+            elif path == "/api/transcript":
+                self._send_json(HTTPStatus.OK, build_run_state(run_dir))
+            else:
+                self._send_json(HTTPStatus.NOT_FOUND, {"error": "unknown endpoint"})
+
+        def _handle_otp_start(self) -> None:
+            email = str(self._read_body().get("email") or "").strip()
+            if not email:
+                self._send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid_email"})
+                return
+            try:
+                auth.otp_start(email)
+            except auth.RelayError as exc:
+                self._send_relay_error(exc)
+                return
+            self._send_json(HTTPStatus.OK, {"ok": True})
+
+        def _handle_otp_verify(self) -> None:
+            body = self._read_body()
+            email = str(body.get("email") or "").strip()
+            code = str(body.get("code") or "").strip()
+            if not email or not code:
+                self._send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid_code"})
+                return
+            try:
+                result = auth.otp_verify(email, code)
+            except auth.RelayError as exc:
+                self._send_relay_error(exc)
+                return
+            auth.write_auth(
+                email=result.get("email") or email,
+                token=result["token"],
+                verified_at=result.get("expires_at") or "",
+            )
+            verified_email = result.get("email") or email
+            self._send_json(HTTPStatus.OK, {"verified": True, "email": verified_email})
+
+        def _handle_forget(self) -> None:
+            auth.forget()
+            self._send_json(HTTPStatus.OK, {"ok": True})
+
+        def _handle_report_send(self) -> None:
+            if not self._has_session():
+                self._send_json(HTTPStatus.FORBIDDEN, {"error": "forbidden"})
+                return
+            record = auth.read_auth()
+            if record is None:
+                self._send_json(HTTPStatus.UNAUTHORIZED, {"error": "unverified"})
+                return
+            run_param = str(self._read_body().get("run") or "") or None
+            run_dir = resolve_run_dir(state.base_dir, run_param, state.run_dir)
+            if run_dir is None:
+                self._send_json(HTTPStatus.NOT_FOUND, {"error": "unknown run"})
+                return
+
+            from strix.viewer.report_pdf import build_encrypted_report
+
+            pdf_bytes, password, filename = build_encrypted_report(run_dir)
+            summary = read_run_summary(run_dir)
+            run_name = str(summary.get("run_name") or run_dir.name)
+            target = primary_target(summary) or "unknown target"
+            try:
+                # The password is intentionally NOT passed here; only the
+                # encrypted PDF bytes reach the relay.
+                auth.report_send(record["token"], pdf_bytes, filename, run_name, target)
+            except auth.RelayError as exc:
+                self._send_relay_error(exc)
+                return
+            # The password is returned only to the local (127.0.0.1) browser.
+            self._send_json(
+                HTTPStatus.OK,
+                {"ok": True, "password": password, "filename": filename},
+            )
+
+        # Cap on a steering message so a runaway client cannot flood the agent.
+        _STEER_MESSAGE_MAX = 4000
+
+        def _handle_steer(self) -> None:
+            if not self._has_session():
+                self._send_json(HTTPStatus.FORBIDDEN, {"error": "forbidden"})
+                return
+            body = self._read_body()
+            agent_id = body.get("agent_id")
+            message = body.get("message")
+            if not isinstance(agent_id, str) or not agent_id.strip():
+                self._send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid_agent_id"})
+                return
+            if (
+                not isinstance(message, str)
+                or not message.strip()
+                or len(message) > self._STEER_MESSAGE_MAX
+            ):
+                self._send_json(HTTPStatus.BAD_REQUEST, {"error": "invalid_message"})
+                return
+            if state.steer_handler is None:
+                # Standalone / finished-run viewing has no live scan to steer.
+                self._send_json(HTTPStatus.FORBIDDEN, {"error": "steering_unavailable"})
+                return
+            delivered = state.steer_handler(agent_id, message)
+            if delivered:
+                self._send_json(HTTPStatus.OK, {"ok": True})
+            else:
+                self._send_json(HTTPStatus.OK, {"ok": False, "error": "not_delivered"})
+
+        def _send_relay_error(self, exc: auth.RelayError) -> None:
+            status_by_code = {
+                "rate_limited": HTTPStatus.TOO_MANY_REQUESTS,
+                "invalid_email": HTTPStatus.BAD_REQUEST,
+                "work_email_required": HTTPStatus.BAD_REQUEST,
+                "invalid_code": HTTPStatus.FORBIDDEN,
+                "reverify": HTTPStatus.UNAUTHORIZED,
+                "forbidden": HTTPStatus.FORBIDDEN,
+                "too_large": HTTPStatus.REQUEST_ENTITY_TOO_LARGE,
+                "unavailable": HTTPStatus.BAD_GATEWAY,
+            }
+            status = status_by_code.get(exc.code, HTTPStatus.BAD_GATEWAY)
+            self._send_json(status, {"error": exc.code})
+
+        def _cookies(self) -> dict[str, str]:
+            jar: dict[str, str] = {}
+            for chunk in (self.headers.get("Cookie") or "").split(";"):
+                name, sep, value = chunk.strip().partition("=")
+                if sep:
+                    jar[name] = value
+            return jar
+
+        def _has_session(self) -> bool:
+            """True when the request carries this process's session capability.
+
+            The cookie is set only when the SPA is served (index.html), so only
+            the browser this process handed the page to can pass. A direct
+            caller on an exposed port has no cookie and is rejected.
+            """
+            supplied = self._cookies().get(SESSION_COOKIE, "")
+            return bool(supplied) and secrets.compare_digest(supplied, state.session_token)
+
+        def _token_presented(self, query: dict[str, list[str]]) -> bool:
+            """True when the request carries the correct bootstrap token.
+
+            The token reaches the operator's browser through the URL printed /
+            opened by the process that started the server, a channel an
+            arbitrary network caller on an exposed port cannot observe.
+            """
+            supplied = (query.get("token") or [""])[0]
+            return bool(supplied) and secrets.compare_digest(supplied, state.session_token)
+
+        def _handle_static(self, path: str, query: dict[str, list[str]]) -> None:
+            target = self._resolve_asset(path)
+            if target is None:
+                # SPA fallback: unknown non-asset routes render index.html so
+                # client-side deep links work.
+                target = state.assets_dir / "index.html"
+            is_index = target.name == "index.html"
+            if not target.is_file():
+                self._send_json(HTTPStatus.NOT_FOUND, {"error": "not found"})
+                return
+            content = target.read_bytes()
+            content_type, _ = mimetypes.guess_type(str(target))
+            self.send_response(HTTPStatus.OK)
+            self.send_header("Content-Type", content_type or "application/octet-stream")
+            self.send_header("Content-Length", str(len(content)))
+            if is_index and self._token_presented(query):
+                # Exchange the bootstrap token for the per-process session
+                # capability. Issued only when the correct token is presented,
+                # so a caller who merely reaches ``/`` never obtains it.
+                # HttpOnly (JS never needs it; fetch sends it automatically) and
+                # SameSite=Strict (never sent from a cross-site context).
+                self.send_header(
+                    "Set-Cookie",
+                    f"{SESSION_COOKIE}={state.session_token}; Path=/; HttpOnly; SameSite=Strict",
+                )
+            self.end_headers()
+            self.wfile.write(content)
+
+        def _resolve_asset(self, path: str) -> Path | None:
+            rel = unquote(path).lstrip("/")
+            if not rel or rel.endswith("/"):
+                return None
+            root = state.assets_dir.resolve()
+            candidate = (root / rel).resolve()
+            # Path-traversal guard: never serve outside the bundle root.
+            if root != candidate and root not in candidate.parents:
+                logger.warning("viewer rejected traversal attempt: %s", path)
+                return None
+            return candidate if candidate.is_file() else None
+
+        def _send_json(self, status: HTTPStatus, payload: Any) -> None:
+            body = json.dumps(payload).encode("utf-8")
+            self.send_response(status)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(body)))
+            self.end_headers()
+            self.wfile.write(body)
+
+    return ViewerHandler
+
+
+def authorized_url(base_url: str, token: str) -> str:
+    """URL that bootstraps the viewer session for the operator.
+
+    Presenting ``token`` on the initial page load is what mints the session
+    cookie, so this URL is printed / opened only for the operator who started
+    the server. Sharing it (rather than the bare ``base_url``) is what lets a
+    trusted remote user authorize when the viewer is exposed with ``--host``.
+    """
+    return f"{base_url}/?{urlencode({'token': token})}"
+
+
+def serve(
+    run_dir: Path,
+    *,
+    host: str = "127.0.0.1",
+    port: int = 0,
+    open_browser: bool = True,
+    steer_handler: Callable[[str, str], bool] | None = None,
+) -> tuple[ThreadingHTTPServer, str, str]:
+    """Start the viewer server on a background thread; return (server, url, token).
+
+    ``url`` is the bare base; pass it through ``authorized_url(url, token)`` to
+    build the operator link that authorizes the browser.
+
+    Binds an ephemeral port by default. If a fixed ``port`` is requested but in
+    use, falls back to an ephemeral port. Reused by both the ``strix view``
+    command and the in-TUI launcher; callers own the server's lifetime.
+
+    ``steer_handler`` is supplied only by the in-TUI launcher, which runs inside
+    the live scan process and can forward a message to a running agent. Left
+    ``None`` (standalone ``strix view``), steering is reported unavailable.
+    """
+    assets_dir = bundle_dir()
+    state = _ViewerState(run_dir=run_dir, assets_dir=assets_dir, steer_handler=steer_handler)
+    handler = _make_handler(state)
+
+    try:
+        httpd = ThreadingHTTPServer((host, port), handler)
+    except OSError:
+        if port == 0:
+            raise
+        logger.info("viewer port %s unavailable, falling back to an ephemeral port", port)
+        httpd = ThreadingHTTPServer((host, 0), handler)
+
+    httpd.daemon_threads = True
+    bound_port = int(httpd.server_address[1])
+    url = f"http://{host}:{bound_port}"
+
+    thread = threading.Thread(target=httpd.serve_forever, name="strix-viewer", daemon=True)
+    thread.start()
+
+    if open_browser:
+        _open_browser(authorized_url(url, state.session_token))
+
+    return httpd, url, state.session_token
+
+
+def _open_browser(url: str) -> None:
+    try:
+        webbrowser.open(url)
+    except Exception:  # noqa: BLE001 - launching the browser is best-effort
+        logger.debug("could not open browser for %s", url, exc_info=True)
+
+
+__all__ = ["authorized_url", "bundle_dir", "bundle_is_built", "serve"]
@@ -0,0 +1,106 @@
+"""Build the JSON payloads the viewer SPA consumes from a run directory."""
+
+from __future__ import annotations
+
+import json
+import logging
+from typing import TYPE_CHECKING, Any
+
+from strix.core.paths import run_record_path
+
+
+if TYPE_CHECKING:
+    from pathlib import Path
+
+
+logger = logging.getLogger(__name__)
+
+_TERMINAL_STATUSES = {"completed", "stopped", "failed", "interrupted"}
+
+_KNOWN_SEVERITIES = ("critical", "high", "medium", "low")
+
+
+def severity_counts(vulns: list[Any]) -> dict[str, int]:
+    """Bucket vulnerabilities into critical/high/medium/low counts.
+
+    Mirrors the SPA's ``severityCounts``: severities are lowercased and
+    trimmed, and anything outside the four known buckets (``info``,
+    ``informational``, ``unknown``, missing, ...) folds into ``low`` so the
+    shared UI renders cleanly.
+    """
+    counts = dict.fromkeys(_KNOWN_SEVERITIES, 0)
+    for vuln in vulns:
+        raw = vuln.get("severity") if isinstance(vuln, dict) else None
+        severity = str(raw or "").lower().strip()
+        if severity not in counts:
+            severity = "low"
+        counts[severity] += 1
+    return counts
+
+
+def build_run_state(run_dir: Path) -> dict[str, Any]:
+    """Agent graph + full per-agent event/message stream.
+
+    Reuses the Textual-free ``TuiLiveView`` projection so the viewer and the TUI
+    share one parser for ``agents.json`` + ``agents.db`` and never drift.
+    """
+    # Imported lazily so importing strix.viewer does not eagerly pull the TUI.
+    from strix.interface.tui.live_view import TuiLiveView  # noqa: PLC0415
+
+    view = TuiLiveView()
+    view.hydrate_from_run_dir(run_dir)
+    return {"agents": list(view.agents.values()), "events": view.events}
+
+
+def read_run_summary(run_dir: Path) -> dict[str, Any]:
+    """The ``run.json`` record plus a computed ``finished`` flag."""
+    record = _load_json(run_record_path(run_dir), default={})
+    if not isinstance(record, dict):
+        record = {}
+    status = record.get("status")
+    finished = status in _TERMINAL_STATUSES and bool(record.get("end_time"))
+    return {**record, "finished": finished}
+
+
+def primary_target(record: dict[str, Any]) -> str | None:
+    """The first target's original string from a run record, or None."""
+    targets = record.get("targets_info")
+    if isinstance(targets, list):
+        for entry in targets:
+            if isinstance(entry, dict):
+                original = entry.get("original")
+                if isinstance(original, str) and original:
+                    return original
+    return None
+
+
+def read_vulnerabilities(run_dir: Path) -> list[Any]:
+    """The ``vulnerabilities.json`` list (empty until a scan writes it)."""
+    data = _load_json(run_dir / "vulnerabilities.json", default=[])
+    return data if isinstance(data, list) else []
+
+
+def read_report_markdown(run_dir: Path) -> str:
+    """The executive report markdown (empty until a scan writes it)."""
+    report_path = run_dir / "penetration_test_report.md"
+    try:
+        return report_path.read_text(encoding="utf-8")
+    except OSError:
+        return ""
+
+
+def _load_json(path: Path, *, default: Any) -> Any:
+    try:
+        return json.loads(path.read_text(encoding="utf-8"))
+    except (OSError, json.JSONDecodeError):
+        return default
+
+
+__all__ = [
+    "build_run_state",
+    "primary_target",
+    "read_report_markdown",
+    "read_run_summary",
+    "read_vulnerabilities",
+    "severity_counts",
+]
@@ -0,0 +1,15 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/png" href="./logo.png" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <meta name="color-scheme" content="dark" />
+    <title>Strix Results</title>
+    <script type="module" crossorigin src="./assets/index-C6BkTz7a.js"></script>
+    <link rel="stylesheet" crossorigin href="./assets/index-C0NveaV7.css">
+  </head>
+  <body>
+    <div id="root"></div>
+  </body>
+</html>
@@ -0,0 +1,14 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/png" href="./logo.png" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <meta name="color-scheme" content="dark" />
+    <title>Strix Results</title>
+  </head>
+  <body>
+    <div id="root"></div>
+    <script type="module" src="/src/main.tsx"></script>
+  </body>
+</html>
@@ -0,0 +1,34 @@
+{
+  "name": "strix-viewer",
+  "private": true,
+  "version": "0.0.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "vite build",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "@dagrejs/dagre": "^2.0.4",
+    "@xyflow/react": "^12.10.1",
+    "clsx": "^2.1.1",
+    "diff": "^8.0.3",
+    "highlight.js": "^11.11.1",
+    "lucide-react": "^0.563.0",
+    "react": "^19.2.0",
+    "react-dom": "^19.2.0",
+    "react-markdown": "^10.1.0",
+    "remark-gfm": "^4.0.1",
+    "tailwind-merge": "^3.4.0"
+  },
+  "devDependencies": {
+    "@tailwindcss/vite": "^4.1.0",
+    "@types/diff": "^7.0.2",
+    "@types/react": "^19.2.0",
+    "@types/react-dom": "^19.2.0",
+    "@vitejs/plugin-react": "^5.0.0",
+    "tailwindcss": "^4.1.0",
+    "typescript": "^5.6.0",
+    "vite": "^6.0.0"
+  }
+}
@@ -0,0 +1,805 @@
+import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+import {
+  ArrowLeft,
+  AlertCircle,
+  Waypoints,
+  Mail,
+  ChevronDown,
+  Wrench,
+  FileCheck2,
+  CalendarClock,
+  Radar,
+  GitPullRequest,
+  Rocket,
+  ArrowUpRight,
+  History,
+} from "lucide-react";
+import type { Vulnerability, VulnerabilitySeverity } from "@/types/issues";
+import { SEVERITY_COLORS } from "@/types/issues";
+import { getSeverityDot } from "@/lib/vulnerability-utils";
+import VulnerabilityDetail from "@/components/vulnerability/VulnerabilityDetail";
+import { ContentSection } from "@/components/vulnerability/ContentSection";
+import { IssueSeveritySummary } from "@/components/IssueSeveritySummary";
+import AgentGraph from "@/components/live/AgentGraph";
+import { buildGraphAgents } from "@/components/live/AgentTranscript";
+import AgentDetailModal from "@/components/live/AgentDetailModal";
+import { ScanPromptComposer } from "@/components/live/ScanPromptComposer";
+import { severityCounts, type ParsedRunSummary } from "@/lib/local-run-parser";
+import {
+  fetchAll,
+  fetchAuthStatus,
+  fetchCapabilities,
+  fetchRunSummary,
+  fetchRuns,
+  fetchTranscript,
+  fetchVulnerabilities,
+  forgetAuth,
+  type AuthStatus,
+  type LoadedRun,
+  type RunsPayload,
+} from "@/data/serverSource";
+import { SIGNUP_URL, ctaUrl, trackCta } from "@/lib/cta";
+import { runTitle } from "@/lib/target-utils";
+import Sidebar from "@/components/Sidebar";
+import PastRunsView from "@/components/PastRunsView";
+import EmailReportView from "@/components/EmailReportView";
+import { RunDetails } from "@/components/RunDetails";
+import { TrustToast } from "@/components/TrustToast";
+import FeatureDetail from "@/components/FeatureDetail";
+import { ProTile, ProInlineCta, type ProItem } from "@/components/ProCta";
+import { FEATURES } from "@/lib/pro-features";
+
+export type View = "overview" | "issues" | "agents" | "history" | "feature" | "email";
+
+const TRUST_BANNER =
+  "Your findings stay on your machine. They're rendered here locally in your browser and never uploaded or stored by Strix. Emailing a report is an explicit opt-in that sends an encrypted copy only you can open.";
+
+const SEVERITY_ORDER: VulnerabilitySeverity[] = ["critical", "high", "medium", "low"];
+const POLL_MS = 500;
+
+// Curated inline CTAs. Continuous-coverage row on Overview (the restyled upsell
+// tiles), plus the recommendations pairing.
+const RECOMMENDATION_CTAS: ProItem[] = [
+  { title: "One-click autofix + open a fix PR", desc: "Fix it for you and open a PR, retested.", slug: "autofix", icon: Wrench },
+  { title: "Export SOC 2 / ISO 27001 report", desc: "Share an auditor-ready report with your team.", slug: "compliance", icon: FileCheck2 },
+];
+const COVERAGE_CTAS: ProItem[] = [
+  { title: "Scheduled pentesting", desc: "Continuous coverage for your whole org.", slug: "scheduled", icon: CalendarClock },
+  { title: "Attack surface monitoring", desc: "Continuous coverage for your whole org.", slug: "asm", icon: Radar },
+  { title: "PR reviews", desc: "Pentest every pull request your team opens.", slug: "pr_reviews", icon: GitPullRequest },
+];
+
+export default function App() {
+  const [activeRun, setActiveRun] = useState<string | null>(null);
+  const [run, setRun] = useState<LoadedRun | null>(null);
+  const [error, setError] = useState<string | null>(null);
+  const [selectedId, setSelectedId] = useState<string | null>(null);
+  const [view, setView] = useState<View>("overview");
+  const [activeFeature, setActiveFeature] = useState<string | null>(null);
+  const [auth, setAuth] = useState<AuthStatus | null>(null);
+  const [runs, setRuns] = useState<RunsPayload | null>(null);
+  const [emailPurpose, setEmailPurpose] = useState<"report" | "verify">("report");
+  const [emailSkipDisclosure, setEmailSkipDisclosure] = useState(false);
+  // Whether this viewer can steer a live scan (true only inside the in-TUI
+  // launcher that shares the running scan's coordinator + event loop).
+  const [canSteer, setCanSteer] = useState(false);
+
+  const refreshAuth = useCallback(async () => {
+    try {
+      setAuth(await fetchAuthStatus());
+    } catch {
+      /* auth status is best-effort; the launched run stays viewable */
+    }
+  }, []);
+
+  const refreshRuns = useCallback(async () => {
+    try {
+      setRuns(await fetchRuns());
+    } catch {
+      /* history list is best-effort */
+    }
+  }, []);
+
+  useEffect(() => {
+    void refreshAuth();
+    void refreshRuns();
+    // Capabilities never change over a session, so fetch once on mount.
+    fetchCapabilities()
+      .then((caps) => setCanSteer(caps.can_steer))
+      .catch(() => {
+        /* absence of steering is the safe default */
+      });
+  }, [refreshAuth, refreshRuns]);
+
+  // Live polling, scoped to the active run. Re-runs when the active run changes
+  // so switching to a past run (?run=<name>) reloads its data; a finished run
+  // does a single full fetch and stops.
+  const finishedRef = useRef(false);
+  useEffect(() => {
+    let cancelled = false;
+    let timer: ReturnType<typeof setTimeout> | undefined;
+    finishedRef.current = false;
+
+    const schedule = () => {
+      timer = setTimeout(tick, POLL_MS);
+    };
+
+    const tick = async () => {
+      if (cancelled) return;
+      try {
+        const { summary, raw, finished } = await fetchRunSummary(activeRun);
+        if (cancelled) return;
+        if (finished && !finishedRef.current) {
+          finishedRef.current = true;
+          const full = await fetchAll(activeRun);
+          if (!cancelled) setRun(full);
+          return; // stop polling
+        }
+        const [transcript, vulnerabilities] = await Promise.all([
+          fetchTranscript(activeRun).catch(() => ({ agents: [], events: [] })),
+          fetchVulnerabilities(summary.runId, activeRun).catch(() => [] as Vulnerability[]),
+        ]);
+        if (cancelled) return;
+        setRun((prev) => ({
+          summary,
+          raw,
+          finished,
+          transcript,
+          vulnerabilities,
+          reportMarkdown: prev?.reportMarkdown ?? null,
+        }));
+        schedule();
+      } catch (e) {
+        if (cancelled) return;
+        setError(e instanceof Error ? e.message : "Could not load run data.");
+        schedule();
+      }
+    };
+
+    (async () => {
+      try {
+        const full = await fetchAll(activeRun);
+        if (cancelled) return;
+        setRun(full);
+        if (full.finished) {
+          finishedRef.current = true;
+        } else {
+          schedule();
+        }
+      } catch (e) {
+        if (cancelled) return;
+        setError(e instanceof Error ? e.message : "Could not load run data.");
+        schedule();
+      }
+    })();
+
+    return () => {
+      cancelled = true;
+      if (timer) clearTimeout(timer);
+    };
+  }, [activeRun]);
+
+  const counts = useMemo(
+    () => (run ? severityCounts(run.vulnerabilities) : null),
+    [run]
+  );
+  const selected = run?.vulnerabilities.find((v) => v.id === selectedId) ?? null;
+  const agentCount = run?.transcript.agents.length ?? 0;
+  const verified = auth?.verified === true;
+
+  // Per-run guard for the default view: land on Agents while a scan is live,
+  // Overview once it finishes. Applied at most once per run and never once the
+  // user has navigated manually (userSetView flips the guard).
+  const initialViewAppliedRef = useRef(false);
+
+  // Reset the guard whenever the active run changes so the newly selected run
+  // gets its own default.
+  useEffect(() => {
+    initialViewAppliedRef.current = false;
+  }, [activeRun]);
+
+  useEffect(() => {
+    if (initialViewAppliedRef.current || !run) return;
+    if (run.finished) {
+      initialViewAppliedRef.current = true;
+      setView("overview");
+    } else if (agentCount > 0) {
+      // Live and agents have appeared: default to the agent graph. If it is
+      // live but no agents exist yet, wait (do not apply, do not set the flag).
+      initialViewAppliedRef.current = true;
+      setView("agents");
+    }
+  }, [run, agentCount]);
+
+  // User-initiated navigation: mark the default guard applied so the per-run
+  // default effect never yanks the user off the view they chose.
+  const userSetView = useCallback((v: View) => {
+    initialViewAppliedRef.current = true;
+    setView(v);
+  }, []);
+
+  const selectRun = useCallback((name: string) => {
+    setActiveRun(name);
+    setSelectedId(null);
+    setRun(null);
+    setError(null);
+    // Reset the guard so the per-run default applies to the newly selected run.
+    initialViewAppliedRef.current = false;
+  }, []);
+
+  const goEmail = useCallback((skipDisclosure: boolean, surface: string) => {
+    trackCta("email_report", surface);
+    setEmailPurpose("report");
+    setEmailSkipDisclosure(skipDisclosure);
+    userSetView("email");
+  }, [userSetView]);
+
+  // Sidebar entry keeps the disclosure (first place those users see it);
+  const openEmail = useCallback(() => goEmail(false, "sidebar"), [goEmail]);
+  // the Overview CTA already states the tradeoff, so it starts the flow directly.
+  const openEmailFromOverview = useCallback(() => goEmail(true, "overview"), [goEmail]);
+
+  const openHistory = useCallback(() => {
+    void refreshRuns();
+    userSetView("history");
+  }, [refreshRuns, userSetView]);
+
+  const onPastRunsVerified = useCallback(async () => {
+    await refreshAuth();
+    await refreshRuns();
+  }, [refreshAuth, refreshRuns]);
+
+  const selectFeature = useCallback((slug: string) => {
+    trackCta(slug, "sidebar_nav");
+    setActiveFeature(slug);
+    userSetView("feature");
+  }, [userSetView]);
+
+  const onForget = useCallback(async () => {
+    await forgetAuth();
+    await refreshAuth();
+    await refreshRuns();
+  }, [refreshAuth, refreshRuns]);
+
+  return (
+    <div className="min-h-screen bg-black text-white flex">
+      <Sidebar
+        view={view}
+        onSelectView={(v) => {
+          if (v === "history") openHistory();
+          else userSetView(v);
+        }}
+        activeFeature={activeFeature}
+        onSelectFeature={selectFeature}
+        issuesCount={run?.vulnerabilities.length ?? 0}
+        agentCount={agentCount}
+        runCount={runs?.count ?? 0}
+        verified={verified}
+        email={auth?.email ?? null}
+        onOpenEmail={openEmail}
+        onOpenHistory={openHistory}
+        onForget={() => void onForget()}
+      />
+
+      <div className="flex-1 min-w-0">
+        {/* Top bar */}
+        <div className="border-b border-[#222]">
+          <div className="max-w-[72rem] mx-auto px-6 py-4 flex items-center gap-1.5">
+            <a
+              href={ctaUrl("https://app.strix.ai", "logo")}
+              target="_blank"
+              rel="noopener noreferrer"
+              onClick={() => trackCta("logo", "topbar")}
+              className="flex items-center gap-1.5 opacity-90 transition-opacity hover:opacity-100 lg:hidden"
+              title="Open Strix Cloud"
+            >
+              <img src="./logo.png" alt="Strix" className="w-10 h-8 object-cover" />
+              <div className="text-base text-white font-medium tracking-tight">Strix</div>
+            </a>
+            <span className="text-xs text-[#666]">Local results</span>
+            {run && <LiveIndicator finished={run.finished} />}
+            <div className="ml-auto flex items-center gap-3">
+              {verified && runs && !runs.locked && runs.runs.length > 0 && (
+                <RunSwitcher
+                  runs={runs}
+                  activeRun={activeRun}
+                  launchedName={runTitle(run?.summary.targets[0] ?? null, run?.summary.runName ?? run?.summary.runId ?? "Current run")}
+                  onSelect={selectRun}
+                />
+              )}
+              <a
+                href={ctaUrl(SIGNUP_URL, "run_in_cloud")}
+                target="_blank"
+                rel="noopener noreferrer"
+                onClick={() => trackCta("run_in_cloud", "topbar")}
+                className="inline-flex items-center gap-1 rounded-lg bg-white px-3 py-1.5 text-xs font-semibold text-black transition-opacity hover:opacity-90"
+              >
+                Run in the cloud
+                <ArrowUpRight className="w-3 h-3" aria-hidden="true" />
+              </a>
+            </div>
+          </div>
+        </div>
+
+        <div className="max-w-[72rem] mx-auto px-6 py-8 space-y-6">
+          {error && !run && view !== "history" && view !== "email" && view !== "feature" && (
+            <div className="rounded-lg px-4 py-3 flex gap-3 items-start border border-red-500/30 bg-red-500/5">
+              <AlertCircle className="w-5 h-5 flex-shrink-0 mt-0.5 text-red-400" aria-hidden="true" />
+              <p className="text-sm text-red-300">{error}</p>
+            </div>
+          )}
+
+          {view === "email" ? (
+            <EmailReportView
+              activeRun={activeRun}
+              auth={auth}
+              purpose={emailPurpose}
+              skipDisclosure={emailSkipDisclosure}
+              onAuthChanged={() => {
+                void refreshAuth();
+                void refreshRuns();
+              }}
+              onExit={(dest) => setView(dest === "history" ? "history" : "overview")}
+            />
+          ) : view === "feature" && activeFeature && FEATURES[activeFeature] ? (
+            <FeatureDetail feature={FEATURES[activeFeature]} />
+          ) : view === "history" ? (
+            <div className="space-y-4">
+              <div className="flex items-center gap-2">
+                <History className="w-5 h-5 text-[#888]" aria-hidden="true" />
+                <h1 className="text-2xl font-semibold text-white">Past runs</h1>
+              </div>
+              <PastRunsView
+                runs={runs}
+                activeRun={activeRun}
+                onSelectRun={selectRun}
+                onVerified={() => void onPastRunsVerified()}
+              />
+            </div>
+          ) : !run && !error ? (
+            <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-10 text-center">
+              <div className="w-6 h-6 mx-auto mb-3 rounded-full border-2 border-[#333] border-t-white animate-spin" />
+              <p className="text-sm text-[#888]">Loading run data…</p>
+            </div>
+          ) : run && counts ? (
+            <>
+              <SummaryHeader summary={run.summary} />
+
+              {/* Tab strip: shown on small screens where the sidebar is hidden. */}
+              <div className="flex gap-5 border-b border-[#2a2a2a] lg:hidden">
+                <TabButton active={view === "overview"} onClick={() => userSetView("overview")}>
+                  Overview
+                </TabButton>
+                <TabButton active={view === "issues"} onClick={() => userSetView("issues")}>
+                  Issues{run.vulnerabilities.length > 0 ? ` (${run.vulnerabilities.length})` : ""}
+                </TabButton>
+                {agentCount > 0 && (
+                  <TabButton active={view === "agents"} onClick={() => userSetView("agents")}>
+                    Agents ({agentCount})
+                  </TabButton>
+                )}
+              </div>
+
+              {view === "overview" ? (
+                <OverviewTab
+                  summary={run.summary}
+                  counts={counts}
+                  total={run.vulnerabilities.length}
+                  reportMarkdown={run.reportMarkdown}
+                  raw={run.raw}
+                  onOpenEmail={openEmailFromOverview}
+                />
+              ) : view === "agents" && agentCount > 0 ? (
+                <AgentsTab run={run} canSteer={canSteer} />
+              ) : selected ? (
+                <div className="space-y-4">
+                  <button
+                    onClick={() => setSelectedId(null)}
+                    className="cursor-pointer inline-flex items-center gap-1.5 text-sm text-[#888] hover:text-white transition-colors"
+                  >
+                    <ArrowLeft className="w-4 h-4" /> Back to all findings
+                  </button>
+                  <VulnerabilityDetail vulnerability={selected} />
+                </div>
+              ) : (
+                <FindingsList
+                  vulnerabilities={run.vulnerabilities}
+                  finished={run.finished}
+                  onSelect={(id) => setSelectedId(id)}
+                />
+              )}
+            </>
+          ) : null}
+        </div>
+      </div>
+      <TrustToast message={TRUST_BANNER} />
+    </div>
+  );
+}
+
+function RunSwitcher({
+  runs,
+  activeRun,
+  launchedName,
+  onSelect,
+}: {
+  runs: RunsPayload;
+  activeRun: string | null;
+  launchedName: string;
+  onSelect: (name: string) => void;
+}) {
+  const [open, setOpen] = useState(false);
+  const activeEntry = runs.runs.find((r) => r.name === activeRun);
+  const current = activeEntry ? runTitle(activeEntry.target, activeEntry.name) : launchedName;
+  return (
+    <div className="relative">
+      <button
+        onClick={() => setOpen((o) => !o)}
+        onBlur={() => setTimeout(() => setOpen(false), 150)}
+        className="flex items-center gap-1.5 rounded-lg px-2.5 py-1.5 text-xs text-[#aaa] transition-colors hover:text-white"
+        style={{ border: "1px solid #2a2a2a" }}
+      >
+        <History className="w-3.5 h-3.5" aria-hidden="true" />
+        <span className="max-w-[160px] truncate">{current}</span>
+        <ChevronDown className="w-3.5 h-3.5" aria-hidden="true" />
+      </button>
+      {open && (
+        <div
+          className="absolute right-0 z-50 mt-1.5 max-h-80 w-64 overflow-y-auto rounded-lg py-1 shadow-xl"
+          style={{ border: "1px solid #2a2a2a", background: "#0a0a0a" }}
+        >
+          {runs.runs.map((r) => {
+            const active = r.name === activeRun;
+            return (
+              <button
+                key={r.name}
+                onMouseDown={() => onSelect(r.name)}
+                className={`flex w-full items-center gap-2 px-3 py-2 text-left text-xs transition-colors hover:bg-[rgba(255,255,255,0.06)] ${
+                  active ? "text-white" : "text-[#aaa]"
+                }`}
+              >
+                <span className="min-w-0 flex-1">
+                  <span className="block truncate">{runTitle(r.target, r.name)}</span>
+                  {r.target && <span className="block truncate font-mono text-[#666]">{r.target}</span>}
+                </span>
+                {active && <span className="h-1.5 w-1.5 flex-shrink-0 rounded-full bg-emerald-400" />}
+              </button>
+            );
+          })}
+        </div>
+      )}
+    </div>
+  );
+}
+
+function LiveIndicator({ finished }: { finished: boolean }) {
+  if (finished) {
+    return (
+      <span className="ml-3 inline-flex items-center gap-1.5 text-xs text-[#888]">
+        <span className="w-1.5 h-1.5 rounded-full bg-[#555]" />
+        Complete
+      </span>
+    );
+  }
+  return (
+    <span className="ml-3 inline-flex items-center gap-1.5 text-xs text-emerald-400">
+      <span className="relative flex h-1.5 w-1.5">
+        <span className="absolute inline-flex h-full w-full rounded-full bg-emerald-400 opacity-75 animate-ping" />
+        <span className="relative inline-flex h-1.5 w-1.5 rounded-full bg-emerald-400" />
+      </span>
+      Live
+    </span>
+  );
+}
+
+function formatDuration(seconds: number | null): string | null {
+  if (seconds == null) return null;
+  if (seconds < 60) return `${seconds}s`;
+  const m = Math.floor(seconds / 60);
+  if (m < 60) return `${m}m`;
+  const h = Math.floor(m / 60);
+  return `${h}h ${m % 60}m`;
+}
+
+function SummaryHeader({ summary }: { summary: ParsedRunSummary }) {
+  const duration = formatDuration(summary.durationSeconds);
+  return (
+    <div>
+      <h1 className="text-2xl font-semibold text-white">
+        {runTitle(summary.targets[0] ?? null, summary.runName ?? summary.runId ?? "Scan results")}
+      </h1>
+      <div className="mt-1 flex flex-wrap items-center gap-x-3 gap-y-1 text-sm text-[#888]">
+        {summary.targets.length > 0 && (
+          <span className="font-mono text-[#aaa]">{summary.targets.join(", ")}</span>
+        )}
+        {summary.scanMode && <Meta label={summary.scanMode} />}
+        {duration && <Meta label={duration} />}
+        {summary.status && <Meta label={summary.status} />}
+      </div>
+    </div>
+  );
+}
+
+function Meta({ label }: { label: string }) {
+  return (
+    <>
+      <span className="text-[#333]">·</span>
+      <span className="capitalize">{label}</span>
+    </>
+  );
+}
+
+function FindingsList({
+  vulnerabilities,
+  finished,
+  onSelect,
+}: {
+  vulnerabilities: Vulnerability[];
+  finished: boolean;
+  onSelect: (id: string) => void;
+}) {
+  const sorted = [...vulnerabilities].sort(
+    (a, b) => SEVERITY_ORDER.indexOf(a.severity) - SEVERITY_ORDER.indexOf(b.severity)
+  );
+  if (sorted.length === 0) {
+    return (
+      <div className="space-y-4">
+        <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-8 text-center text-sm text-[#888]">
+          {finished ? "No findings in this run." : "No findings yet. The scan is still running…"}
+        </div>
+        {finished && (
+          <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+            <p className="text-sm font-medium text-white">Stay ahead of new exposures</p>
+            <p className="mt-0.5 mb-3 text-xs text-[#666]">
+              Attack surface monitoring catches new exposures for your org over time.
+            </p>
+            <ProInlineCta
+              label="Attack surface monitoring"
+              desc="Continuous coverage for your whole org."
+              slug="asm"
+              surface="empty_state"
+              icon={Radar}
+            />
+          </div>
+        )}
+      </div>
+    );
+  }
+  return (
+    <div className="space-y-2">
+      {sorted.map((v) => (
+        <button
+          key={v.id}
+          onClick={() => onSelect(v.id)}
+          className="cursor-pointer w-full text-left rounded-lg border border-[#222] hover:border-[#444] bg-[rgba(255,255,255,0.02)] px-4 py-3 transition-colors flex items-center gap-3"
+        >
+          <span className={`w-2.5 h-2.5 rounded-full flex-shrink-0 ${getSeverityDot(v.severity)}`} aria-hidden="true" />
+          <span className="flex-1 min-w-0">
+            <span className="block text-sm font-medium text-white truncate">{v.title}</span>
+            {v.target && (
+              <span className="block text-xs text-[#666] font-mono truncate">{v.target}</span>
+            )}
+          </span>
+          <span
+            className={`text-xs font-semibold px-2 py-0.5 rounded-full border capitalize ${SEVERITY_COLORS[v.severity]}`}
+          >
+            {v.severity}
+          </span>
+        </button>
+      ))}
+    </div>
+  );
+}
+
+/** Strip a single leading markdown heading (report sections embed their own). */
+function stripLeadingHeading(md: string): string {
+  return md.replace(/^\s*#{1,6}[ \t]+.*(?:\r?\n)+/, "").trimStart();
+}
+
+function dedupeHeadings(md: string): string {
+  const out: string[] = [];
+  let lastHeading: string | null = null;
+  for (const line of md.split("\n")) {
+    const m = line.match(/^#{1,6}\s+(.*)$/);
+    if (m) {
+      const norm = m[1].trim().toLowerCase();
+      if (norm === lastHeading) continue;
+      lastHeading = norm;
+    } else if (line.trim() !== "") {
+      lastHeading = null;
+    }
+    out.push(line);
+  }
+  return out.join("\n");
+}
+
+/** Primary local CTA: email an encrypted PDF. Verify-email affordance, no lock. */
+function EmailReportCta({ onOpenEmail }: { onOpenEmail: () => void }) {
+  return (
+    <button
+      onClick={onOpenEmail}
+      className="group w-full cursor-pointer rounded-xl border border-emerald-500/25 bg-emerald-500/[0.06] p-4 text-left transition-colors hover:border-emerald-500/40"
+    >
+      <div className="flex items-center gap-3">
+        <div
+          className="flex h-9 w-9 flex-shrink-0 items-center justify-center rounded-lg"
+          style={{ border: "1px solid rgba(16,185,129,0.3)", background: "rgba(16,185,129,0.08)" }}
+        >
+          <Mail className="h-4 w-4 text-emerald-400" aria-hidden="true" />
+        </div>
+        <div className="min-w-0 flex-1">
+          <p className="text-sm font-semibold text-white">Email an encrypted PDF report of this run</p>
+          <p className="mt-0.5 text-xs text-[#888]">
+            Encrypted with a key only you can see, email verified with a one-time code before sending.
+          </p>
+        </div>
+        <span className="flex-shrink-0 rounded-lg bg-white px-3 py-1.5 text-xs font-semibold text-black transition-opacity group-hover:opacity-90">
+          Email report
+        </span>
+      </div>
+    </button>
+  );
+}
+
+function OverviewTab({
+  summary,
+  counts,
+  total,
+  reportMarkdown,
+  raw,
+  onOpenEmail,
+}: {
+  summary: ParsedRunSummary;
+  counts: Record<VulnerabilitySeverity, number>;
+  total: number;
+  reportMarkdown: string | null;
+  raw: Record<string, unknown>;
+  onOpenEmail: () => void;
+}) {
+  const sections = (
+    [
+      ["Executive Summary", summary.executiveSummary],
+      ["Technical Analysis", summary.technicalAnalysis],
+      ["Methodology", summary.methodology],
+      ["Recommendations", summary.recommendations],
+    ] as const
+  )
+    .filter(([, content]) => !!content)
+    .map(([title, content]) => ({ title, content: stripLeadingHeading(content as string) }));
+
+  return (
+    <div className="space-y-6">
+      <RunDetails raw={raw} durationSeconds={summary.durationSeconds} />
+
+      {total > 0 && (
+        <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+          <IssueSeveritySummary findings={{ total, ...counts }} />
+        </div>
+      )}
+
+      {/* Primary CTA: the one primary on Overview. */}
+      <EmailReportCta onOpenEmail={onOpenEmail} />
+
+      {sections.length > 0 ? (
+        <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5 space-y-8">
+          {sections.map((s) => (
+            <ContentSection key={s.title} title={s.title} content={s.content} />
+          ))}
+        </div>
+      ) : reportMarkdown ? (
+        <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+          <ContentSection content={dedupeHeadings(reportMarkdown)} />
+        </div>
+      ) : (
+        total === 0 && (
+          <p className="text-sm text-[#888]">No summary available for this run yet.</p>
+        )
+      )}
+
+      {/* Near Recommendations: act on the fixes. */}
+      <div className="grid grid-cols-1 sm:grid-cols-2 gap-3">
+        {RECOMMENDATION_CTAS.map((item) => (
+          <ProTile key={item.slug} item={item} surface="overview" />
+        ))}
+      </div>
+
+      {/* Continuous coverage for your org (restyled upsell tiles). */}
+      <div>
+        <p className="mb-2 text-sm font-semibold text-white">Continuous coverage for your org</p>
+        <div className="grid grid-cols-1 sm:grid-cols-3 gap-3">
+          {COVERAGE_CTAS.map((item) => (
+            <ProTile key={item.slug} item={item} surface="overview" />
+          ))}
+        </div>
+      </div>
+    </div>
+  );
+}
+
+function TabButton({
+  active,
+  onClick,
+  children,
+}: {
+  active: boolean;
+  onClick: () => void;
+  children: React.ReactNode;
+}) {
+  return (
+    <button
+      onClick={onClick}
+      className={`cursor-pointer relative pb-2.5 text-sm font-semibold transition-colors ${
+        active ? "text-white" : "text-[#666] hover:text-white"
+      }`}
+    >
+      {children}
+      {active && <span className="absolute bottom-0 inset-x-0 h-0.5 bg-white rounded-full" />}
+    </button>
+  );
+}
+
+function AgentsTab({ run, canSteer }: { run: LoadedRun; canSteer: boolean }) {
+  const { agents, events } = run.transcript;
+  const graphAgents = useMemo(() => buildGraphAgents(agents, events), [agents, events]);
+  // Clicking a graph node opens the agent's transcript in a modal (matching the
+  // cloud app); no node selected means no modal.
+  const [selectedId, setSelectedId] = useState<string | null>(null);
+  const selectedAgent = selectedId ? (agents.find((a) => a.id === selectedId) ?? null) : null;
+
+  // Live steering is only possible in-process (canSteer) while the scan runs.
+  const steerable = canSteer && !run.finished;
+
+  return (
+    <div className="space-y-5">
+      <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+        <div className="flex items-center gap-2">
+          <Waypoints className="w-4 h-4 text-[#888]" aria-hidden="true" />
+          <h2 className="text-sm font-semibold text-white">Agent graph</h2>
+          <span className="text-xs text-[#666]">
+            {agents.length} agent{agents.length === 1 ? "" : "s"}
+          </span>
+        </div>
+        <p className="mt-1 mb-4 text-xs text-[#666]">
+          Click an agent to open its full transcript.
+        </p>
+        <div className="h-[480px] rounded-lg border border-[#1a1a1a] overflow-hidden">
+          <AgentGraph
+            agents={graphAgents}
+            selectedAgentId={selectedId}
+            onSelectAgent={(id) => setSelectedId(id)}
+            eventsLoaded
+            eventsEmpty={graphAgents.size === 0}
+            scanCompleted={run.finished}
+          />
+        </div>
+      </div>
+
+      {/* Live steering: only in-process while the scan runs. Otherwise omitted. */}
+      {steerable && <ScanPromptComposer agents={agents} />}
+
+      {/* Re-run always routes to Strix Cloud. */}
+      <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+        <p className="text-sm font-semibold text-white">Run this scan with more depth</p>
+        <p className="mt-0.5 text-xs text-[#666]">Re-run this scan on managed infra in the cloud.</p>
+        <div className="mt-3 flex flex-wrap gap-2.5">
+          <ProInlineCta
+            label="Re-run in Strix Cloud with more depth"
+            desc="Run this scan on managed infra with more depth."
+            slug="live_scan"
+            surface="agents"
+            icon={Rocket}
+          />
+        </div>
+      </div>
+
+      {selectedAgent && (
+        <AgentDetailModal
+          agent={selectedAgent}
+          events={events}
+          steerable={steerable}
+          onClose={() => setSelectedId(null)}
+        />
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,20 @@
+// Extracted ProviderIcon from strix-app's AddRepositoryDialog. The dialog itself
+// (and its next/link dependency) is dropped; the IssueSidebar only needs this SVG
+// switch to badge a finding's source-control provider. Web-app targets resolve to
+// provider === null and never reach here (they render a globe icon instead).
+import { Github, Gitlab } from "lucide-react";
+
+function BitbucketIcon({ className }: { className?: string }) {
+  return (
+    <svg viewBox="0 0 24 24" fill="currentColor" className={className} aria-hidden="true">
+      <path d="M2.65 3a.72.72 0 0 0-.72.83l2.86 17.39a.98.98 0 0 0 .96.82h13.72a.72.72 0 0 0 .72-.6l2.86-17.4A.72.72 0 0 0 22.3 3H2.65Zm12.1 12.53H9.3L8.06 8.9h7.8l-1.11 6.63Z" />
+    </svg>
+  );
+}
+
+export function ProviderIcon({ provider, className }: { provider: string; className?: string }) {
+  const cls = className ?? "w-4 h-4";
+  if (provider === "gitlab") return <Gitlab className={`${cls} text-orange-400`} />;
+  if (provider === "bitbucket") return <BitbucketIcon className={`${cls} text-blue-400`} />;
+  return <Github className={`${cls} text-white`} />;
+}
@@ -0,0 +1,375 @@
+import { useEffect, useRef, useState } from "react";
+import { Mail, ShieldCheck, Lock, Copy, Check, Loader2, AlertCircle, ArrowLeft } from "lucide-react";
+import {
+  otpStart,
+  otpVerify,
+  sendReport,
+  type AuthStatus,
+} from "@/data/serverSource";
+import { track } from "@/lib/cta";
+
+/**
+ * The email-report / email-verification flow rendered as its own page (not a
+ * modal, so it never floats over another surface). Report mode ends in the
+ * one-time password panel; verify mode just confirms the email and returns to
+ * the caller. The page unmounts when you navigate away, so state resets each
+ * time it is opened.
+ */
+
+type Step = "disclosure" | "email" | "code" | "sending" | "password";
+
+interface EmailReportViewProps {
+  activeRun: string | null;
+  auth: AuthStatus | null;
+  purpose: "report" | "verify";
+  /**
+   * Skip the report disclosure and start the flow directly (used by the
+   * Overview CTA, which already states the tradeoff). Unverified users land on
+   * the email step; already-verified users send immediately.
+   */
+  skipDisclosure?: boolean;
+  /** Refresh auth + runs after a successful verify (lifts state to App). */
+  onAuthChanged: () => void;
+  /** Leave this page (report "Done" -> overview; verify success -> history). */
+  onExit: (dest: "overview" | "history") => void;
+}
+
+const OTP_START_ERRORS: Record<string, string> = {
+  work_email_required: "Please use your work email, not a personal one.",
+  rate_limited: "Too many requests. Wait a minute and try again.",
+  invalid_email: "That email does not look right. Check it and try again.",
+  unavailable: "The email service is unavailable right now. Try again shortly.",
+};
+
+const SEND_ERRORS: Record<string, string> = {
+  forbidden: "This email was unsubscribed from Strix, so we cannot send to it.",
+  too_large: "This report is too large to email. Try a smaller run.",
+  unavailable: "The email service is unavailable right now. Try again shortly.",
+};
+
+// A small set of common personal providers for instant client-side feedback.
+// The relay is authoritative (it checks the full free-email-domains list).
+const COMMON_FREE_DOMAINS = new Set([
+  "gmail.com", "googlemail.com", "yahoo.com", "ymail.com", "outlook.com",
+  "hotmail.com", "live.com", "icloud.com", "me.com", "aol.com", "proton.me",
+  "protonmail.com", "gmx.com", "mail.com",
+]);
+
+export default function EmailReportView({
+  activeRun,
+  auth,
+  purpose,
+  skipDisclosure = false,
+  onAuthChanged,
+  onExit,
+}: EmailReportViewProps) {
+  const verified = auth?.verified === true;
+  const verifyOnly = purpose === "verify";
+  // Verify mode (and the Overview CTA, which skips the disclosure) start on the
+  // email step; a verified user who skips the disclosure sends immediately.
+  const [step, setStep] = useState<Step>(() => {
+    if (verifyOnly) return "email";
+    if (skipDisclosure) return verified ? "sending" : "email";
+    return "disclosure";
+  });
+  const [email, setEmail] = useState(auth?.email ?? "");
+  const [code, setCode] = useState("");
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+  const [notice, setNotice] = useState<string | null>(null);
+  const [password, setPassword] = useState("");
+  const [filename, setFilename] = useState("");
+  const [copied, setCopied] = useState(false);
+  const [sentTo, setSentTo] = useState("");
+  const autoSentRef = useRef(false);
+
+  const doSend = async () => {
+    setStep("sending");
+    setError(null);
+    const result = await sendReport(activeRun);
+    if (result.ok) {
+      track("report_sent");
+      setPassword(result.password);
+      setFilename(result.filename);
+      setStep("password");
+      return;
+    }
+    if (result.error === "reverify" || result.error === "unverified") {
+      setNotice("Your verification expired. Enter your email to verify again.");
+      setStep("email");
+      return;
+    }
+    setError(SEND_ERRORS[result.error] ?? "Could not send the report. Try again.");
+    setStep("disclosure");
+  };
+
+  const startFlow = () => {
+    setError(null);
+    setNotice(null);
+    if (verified) void doSend();
+    else setStep("email");
+  };
+
+  // A verified user who skipped the disclosure (Overview CTA) sends on arrival.
+  useEffect(() => {
+    if (!verifyOnly && skipDisclosure && verified && !autoSentRef.current) {
+      autoSentRef.current = true;
+      void doSend();
+    }
+    // Run once on mount; the page remounts fresh each time it is opened.
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, []);
+
+  const submitEmail = async () => {
+    const value = email.trim();
+    if (!value) {
+      setError("Enter your email to continue.");
+      return;
+    }
+    const domain = value.slice(value.lastIndexOf("@") + 1).toLowerCase();
+    if (COMMON_FREE_DOMAINS.has(domain)) {
+      track("work_email_required");
+      setError(OTP_START_ERRORS.work_email_required);
+      return;
+    }
+    setBusy(true);
+    setError(null);
+    const result = await otpStart(value);
+    setBusy(false);
+    if (result.ok) {
+      track("email_submitted", { purpose });
+      setNotice(`We sent a 6-digit code to ${value}.`);
+      setStep("code");
+    } else {
+      if (result.error === "work_email_required") track("work_email_required");
+      setError(OTP_START_ERRORS[result.error] ?? "Could not send a code. Try again.");
+    }
+  };
+
+  const submitCode = async () => {
+    const value = code.trim();
+    if (value.length < 4) {
+      setError("Enter the 6-digit code from your email.");
+      return;
+    }
+    setBusy(true);
+    setError(null);
+    const result = await otpVerify(email.trim(), value);
+    setBusy(false);
+    if (!result.verified) {
+      setError("That code did not match. Check it and try again.");
+      return;
+    }
+    track("email_verified", { purpose });
+    setSentTo(result.email);
+    onAuthChanged();
+    if (verifyOnly) onExit("history");
+    else void doSend();
+  };
+
+  const copyPassword = async () => {
+    try {
+      await navigator.clipboard.writeText(password);
+      setCopied(true);
+      setTimeout(() => setCopied(false), 1500);
+    } catch {
+      /* clipboard may be unavailable; the password is visible to copy manually */
+    }
+  };
+
+  const confirmationEmail = sentTo || auth?.email || email.trim();
+
+  return (
+    <div className="mx-auto max-w-md space-y-4">
+      <button
+        onClick={() => onExit(verifyOnly ? "history" : "overview")}
+        className="cursor-pointer inline-flex items-center gap-1.5 text-sm text-[#888] transition-colors hover:text-white"
+      >
+        <ArrowLeft className="h-4 w-4" />
+        {verifyOnly ? "Back to past runs" : "Back to results"}
+      </button>
+
+      <div className="flex items-center gap-2">
+        <Mail className="h-5 w-5 text-[#888]" aria-hidden="true" />
+        <h1 className="text-2xl font-semibold text-white">
+          {verifyOnly ? "Verify your email" : "Email report"}
+        </h1>
+      </div>
+
+      <div
+        className="w-full rounded-2xl bg-[rgba(255,255,255,0.02)] p-6"
+        style={{ border: "1px solid #2a2a2a" }}
+      >
+        <p className="mb-4 text-xs text-[#666]">
+          {verifyOnly
+            ? "We send a one-time code to confirm it is you."
+            : "Verified by a one-time code sent to your email"}
+        </p>
+
+        {error && (
+          <div className="mb-4 flex items-start gap-2 rounded-lg border border-red-500/30 bg-red-500/5 px-3 py-2">
+            <AlertCircle className="mt-0.5 h-4 w-4 flex-shrink-0 text-red-400" aria-hidden="true" />
+            <p className="text-xs text-red-300">{error}</p>
+          </div>
+        )}
+        {notice && !error && step !== "password" && (
+          <p className="mb-4 text-xs text-[#888]">{notice}</p>
+        )}
+
+        {step === "disclosure" && (
+          <div className="space-y-4">
+            <div
+              className="space-y-2.5 rounded-lg p-3.5"
+              style={{ border: "1px solid #222", background: "rgba(255,255,255,0.02)" }}
+            >
+              <div className="flex items-start gap-2.5">
+                <ShieldCheck className="mt-0.5 h-4 w-4 flex-shrink-0 text-emerald-400" aria-hidden="true" />
+                <p className="text-xs leading-relaxed text-[#aaa]">
+                  Viewing stays local and nothing is uploaded. Emailing is an explicit
+                  opt-in: we send an <span className="text-white">encrypted PDF</span>.
+                </p>
+              </div>
+              <div className="flex items-start gap-2.5">
+                <Lock className="mt-0.5 h-4 w-4 flex-shrink-0 text-[#888]" aria-hidden="true" />
+                <p className="text-xs leading-relaxed text-[#aaa]">
+                  The report is encrypted with a password that only you hold. Strix
+                  cannot read it and never stores it. We collect only your email so we
+                  can send it.
+                </p>
+              </div>
+            </div>
+            <button
+              onClick={startFlow}
+              className="w-full cursor-pointer rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90"
+            >
+              {verified ? "Email me the encrypted PDF" : "Continue with your email"}
+            </button>
+            {verified && auth?.email && (
+              <p className="text-center text-xs text-[#666]">Sending to {auth.email}</p>
+            )}
+          </div>
+        )}
+
+        {step === "email" && (
+          <form
+            className="space-y-4"
+            onSubmit={(e) => {
+              e.preventDefault();
+              void submitEmail();
+            }}
+          >
+            <label className="block">
+              <span className="mb-1.5 block text-xs text-[#888]">Your work email</span>
+              <input
+                type="email"
+                autoFocus
+                value={email}
+                onChange={(e) => setEmail(e.target.value)}
+                placeholder="you@company.com"
+                className="w-full rounded-lg bg-black px-3 py-2.5 text-sm text-white outline-none transition-colors focus:border-[#444]"
+                style={{ border: "1px solid #2a2a2a" }}
+              />
+              <span className="mt-1.5 block text-[11px] text-[#666]">Use your work email.</span>
+            </label>
+            <button
+              type="submit"
+              disabled={busy}
+              className="flex w-full cursor-pointer items-center justify-center gap-2 rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90 disabled:opacity-60"
+            >
+              {busy && <Loader2 className="h-4 w-4 animate-spin" aria-hidden="true" />}
+              Send me a code
+            </button>
+          </form>
+        )}
+
+        {step === "code" && (
+          <form
+            className="space-y-4"
+            onSubmit={(e) => {
+              e.preventDefault();
+              void submitCode();
+            }}
+          >
+            <label className="block">
+              <span className="mb-1.5 block text-xs text-[#888]">6-digit code</span>
+              <input
+                inputMode="numeric"
+                autoFocus
+                value={code}
+                onChange={(e) => setCode(e.target.value.replace(/\D/g, "").slice(0, 6))}
+                placeholder="123456"
+                className="w-full rounded-lg bg-black px-3 py-2.5 text-center text-lg font-mono tracking-[0.4em] text-white outline-none transition-colors focus:border-[#444]"
+                style={{ border: "1px solid #2a2a2a" }}
+              />
+            </label>
+            <button
+              type="submit"
+              disabled={busy}
+              className="flex w-full cursor-pointer items-center justify-center gap-2 rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90 disabled:opacity-60"
+            >
+              {busy && <Loader2 className="h-4 w-4 animate-spin" aria-hidden="true" />}
+              {verifyOnly ? "Verify" : "Verify and send"}
+            </button>
+            <button
+              type="button"
+              onClick={() => {
+                setStep("email");
+                setError(null);
+                setNotice(null);
+              }}
+              className="w-full cursor-pointer text-center text-xs text-[#666] transition-colors hover:text-[#aaa]"
+            >
+              Use a different email
+            </button>
+          </form>
+        )}
+
+        {step === "sending" && (
+          <div className="flex flex-col items-center gap-3 py-8">
+            <Loader2 className="h-6 w-6 animate-spin text-white" aria-hidden="true" />
+            <p className="text-sm text-[#aaa]">Generating and encrypting locally...</p>
+          </div>
+        )}
+
+        {step === "password" && (
+          <div className="space-y-4">
+            <div className="flex items-start gap-2.5 rounded-lg border border-emerald-500/30 bg-emerald-500/5 px-3 py-2.5">
+              <Check className="mt-0.5 h-4 w-4 flex-shrink-0 text-emerald-400" aria-hidden="true" />
+              <p className="text-xs text-emerald-200">
+                Sent to {confirmationEmail}. Open the attached PDF with this password.
+              </p>
+            </div>
+            <div>
+              <span className="mb-1.5 block text-xs text-[#888]">Your one-time password</span>
+              <div
+                className="flex items-center gap-2 rounded-lg bg-black p-3"
+                style={{ border: "1px solid #2a2a2a" }}
+              >
+                <code className="flex-1 break-all font-mono text-base text-white">{password}</code>
+                <button
+                  onClick={copyPassword}
+                  className="flex cursor-pointer items-center gap-1 rounded-md px-2 py-1 text-xs text-[#aaa] transition-colors hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+                  style={{ border: "1px solid #2a2a2a" }}
+                >
+                  {copied ? <Check className="h-3.5 w-3.5" /> : <Copy className="h-3.5 w-3.5" />}
+                  {copied ? "Copied" : "Copy"}
+                </button>
+              </div>
+              <p className="mt-2 text-xs text-[#666]">
+                Save this now. Strix never stores it, so we cannot show it again. File:{" "}
+                <span className="font-mono text-[#888]">{filename}</span>
+              </p>
+            </div>
+            <button
+              onClick={() => onExit("overview")}
+              className="w-full cursor-pointer rounded-lg px-4 py-2.5 text-sm font-semibold text-white transition-colors hover:bg-[rgba(255,255,255,0.06)]"
+              style={{ border: "1px solid #2a2a2a" }}
+            >
+              Done
+            </button>
+          </div>
+        )}
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,163 @@
+import { useState } from "react";
+import { Loader2, AlertCircle } from "lucide-react";
+import { otpStart, otpVerify } from "@/data/serverSource";
+import { track } from "@/lib/cta";
+
+/**
+ * Compact inline email -> 6-digit-code verify flow. Unlike EmailReportView this
+ * has no page chrome, no report send, and no password panel: it just confirms
+ * the email so the past-runs list can unlock in place. On success it calls
+ * `onVerified` (the parent refreshes auth + runs).
+ */
+
+const OTP_START_ERRORS: Record<string, string> = {
+  work_email_required: "Please use your work email, not a personal one.",
+  rate_limited: "Too many requests. Wait a minute and try again.",
+  invalid_email: "That email does not look right. Check it and try again.",
+  unavailable: "The email service is unavailable right now. Try again shortly.",
+};
+
+// A small set of common personal providers for instant client-side feedback.
+// The relay is authoritative (it checks the full free-email-domains list).
+const COMMON_FREE_DOMAINS = new Set([
+  "gmail.com", "googlemail.com", "yahoo.com", "ymail.com", "outlook.com",
+  "hotmail.com", "live.com", "icloud.com", "me.com", "aol.com", "proton.me",
+  "protonmail.com", "gmx.com", "mail.com",
+]);
+
+export default function EmailVerifyInline({ onVerified }: { onVerified: () => void }) {
+  const [step, setStep] = useState<"email" | "code">("email");
+  const [email, setEmail] = useState("");
+  const [code, setCode] = useState("");
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+  const [notice, setNotice] = useState<string | null>(null);
+
+  const submitEmail = async () => {
+    const value = email.trim();
+    if (!value) {
+      setError("Enter your email to continue.");
+      return;
+    }
+    const domain = value.slice(value.lastIndexOf("@") + 1).toLowerCase();
+    if (COMMON_FREE_DOMAINS.has(domain)) {
+      track("work_email_required");
+      setError(OTP_START_ERRORS.work_email_required);
+      return;
+    }
+    setBusy(true);
+    setError(null);
+    const result = await otpStart(value);
+    setBusy(false);
+    if (result.ok) {
+      track("email_submitted", { purpose: "verify" });
+      setNotice(`We sent a 6-digit code to ${value}.`);
+      setStep("code");
+    } else {
+      if (result.error === "work_email_required") track("work_email_required");
+      setError(OTP_START_ERRORS[result.error] ?? "Could not send a code. Try again.");
+    }
+  };
+
+  const submitCode = async () => {
+    const value = code.trim();
+    if (value.length < 4) {
+      setError("Enter the 6-digit code from your email.");
+      return;
+    }
+    setBusy(true);
+    setError(null);
+    const result = await otpVerify(email.trim(), value);
+    setBusy(false);
+    if (!result.verified) {
+      setError("That code did not match. Check it and try again.");
+      return;
+    }
+    track("email_verified", { purpose: "verify" });
+    onVerified();
+  };
+
+  return (
+    <div className="mx-auto mt-5 max-w-sm text-left">
+      {error && (
+        <div className="mb-3 flex items-start gap-2 rounded-lg border border-red-500/30 bg-red-500/5 px-3 py-2">
+          <AlertCircle className="mt-0.5 h-4 w-4 flex-shrink-0 text-red-400" aria-hidden="true" />
+          <p className="text-xs text-red-300">{error}</p>
+        </div>
+      )}
+      {notice && !error && <p className="mb-3 text-xs text-[#888]">{notice}</p>}
+
+      {step === "email" ? (
+        <form
+          className="space-y-3"
+          onSubmit={(e) => {
+            e.preventDefault();
+            void submitEmail();
+          }}
+        >
+          <label className="block">
+            <span className="mb-1.5 block text-xs text-[#888]">Your work email</span>
+            <input
+              type="email"
+              autoFocus
+              value={email}
+              onChange={(e) => setEmail(e.target.value)}
+              placeholder="you@company.com"
+              className="w-full rounded-lg bg-black px-3 py-2.5 text-sm text-white outline-none transition-colors focus:border-[#444]"
+              style={{ border: "1px solid #2a2a2a" }}
+            />
+            <span className="mt-1.5 block text-[11px] text-[#666]">Use your work email.</span>
+          </label>
+          <button
+            type="submit"
+            disabled={busy}
+            className="flex w-full cursor-pointer items-center justify-center gap-2 rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90 disabled:opacity-60"
+          >
+            {busy && <Loader2 className="h-4 w-4 animate-spin" aria-hidden="true" />}
+            Send me a code
+          </button>
+        </form>
+      ) : (
+        <form
+          className="space-y-3"
+          onSubmit={(e) => {
+            e.preventDefault();
+            void submitCode();
+          }}
+        >
+          <label className="block">
+            <span className="mb-1.5 block text-xs text-[#888]">6-digit code</span>
+            <input
+              inputMode="numeric"
+              autoFocus
+              value={code}
+              onChange={(e) => setCode(e.target.value.replace(/\D/g, "").slice(0, 6))}
+              placeholder="123456"
+              className="w-full rounded-lg bg-black px-3 py-2.5 text-center text-lg font-mono tracking-[0.4em] text-white outline-none transition-colors focus:border-[#444]"
+              style={{ border: "1px solid #2a2a2a" }}
+            />
+          </label>
+          <button
+            type="submit"
+            disabled={busy}
+            className="flex w-full cursor-pointer items-center justify-center gap-2 rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90 disabled:opacity-60"
+          >
+            {busy && <Loader2 className="h-4 w-4 animate-spin" aria-hidden="true" />}
+            Verify
+          </button>
+          <button
+            type="button"
+            onClick={() => {
+              setStep("email");
+              setError(null);
+              setNotice(null);
+            }}
+            className="w-full cursor-pointer text-center text-xs text-[#666] transition-colors hover:text-[#aaa]"
+          >
+            Use a different email
+          </button>
+        </form>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,96 @@
+import {
+  CalendarClock,
+  WandSparkles,
+  Puzzle,
+  Users,
+  ArrowUpRight,
+} from "lucide-react";
+import { SIGNUP_URL, PRICING_URL, ctaUrl, trackCta } from "@/lib/cta";
+import type { ProFeature } from "@/lib/pro-features";
+import { ProTag } from "@/components/ProCta";
+
+/**
+ * In-app upsell page for a single platform feature. Modeled on the cloud app's
+ * Networks upsell: a centered bordered card with an icon medallion, tier pill,
+ * headline, one-line description, a shared "Included in Strix Pro" bullet list,
+ * then a primary sign-up CTA and a secondary link to all plans.
+ */
+
+const INCLUDED = [
+  {
+    icon: CalendarClock,
+    text: "Continuous coverage: scheduled pentests and attack surface monitoring",
+  },
+  { icon: WandSparkles, text: "One-click autofix that opens a retested pull request" },
+  { icon: Puzzle, text: "Two-way sync to Jira, Linear, and Slack" },
+  { icon: Users, text: "Your whole team, with roles and shared history" },
+];
+
+export default function FeatureDetail({ feature }: { feature: ProFeature }) {
+  const Icon = feature.icon;
+  return (
+    <div className="mx-auto w-full max-w-lg">
+      <div className="rounded-2xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-8 text-center">
+        <div
+          className="mx-auto flex h-12 w-12 items-center justify-center rounded-xl"
+          style={{ border: "1px solid #2a2a2a", background: "rgba(255,255,255,0.04)" }}
+        >
+          <Icon className="h-5 w-5 text-[#888]" aria-hidden="true" />
+        </div>
+
+        <div className="mt-4 flex justify-center">
+          <ProTag label={feature.tier} />
+        </div>
+
+        <h2 className="mt-3 text-2xl font-semibold text-white">{feature.headline}</h2>
+        <p className="mx-auto mt-2 max-w-md text-sm text-[#888]">{feature.description}</p>
+
+        <div
+          className="mt-6 rounded-xl p-4 text-left"
+          style={{ border: "1px solid #222", background: "rgba(255,255,255,0.02)" }}
+        >
+          <p className="mb-3 text-xs font-semibold uppercase tracking-wide text-[#666]">
+            Included in Strix Pro
+          </p>
+          <ul className="space-y-2.5">
+            {INCLUDED.map((item) => {
+              const BulletIcon = item.icon;
+              return (
+                <li key={item.text} className="flex items-start gap-2.5">
+                  <BulletIcon
+                    className="mt-0.5 h-4 w-4 flex-shrink-0 text-[#888]"
+                    aria-hidden="true"
+                  />
+                  <span className="text-sm text-[#aaa]">{item.text}</span>
+                </li>
+              );
+            })}
+          </ul>
+        </div>
+
+        <div className="mt-6 flex flex-col items-center gap-3">
+          <a
+            href={ctaUrl(SIGNUP_URL, feature.slug)}
+            target="_blank"
+            rel="noopener noreferrer"
+            onClick={() => trackCta(feature.slug, "feature_page")}
+            className="inline-flex w-full items-center justify-center gap-1.5 rounded-lg bg-white px-4 py-2.5 text-sm font-semibold text-black transition-opacity hover:opacity-90"
+          >
+            Start free
+            <ArrowUpRight className="h-3.5 w-3.5" aria-hidden="true" />
+          </a>
+          <a
+            href={ctaUrl(PRICING_URL, feature.slug)}
+            target="_blank"
+            rel="noopener noreferrer"
+            onClick={() => trackCta(feature.slug, "feature_page_plans")}
+            className="inline-flex items-center gap-1 text-xs text-[#888] transition-colors hover:text-white"
+          >
+            View all plans
+            <ArrowUpRight className="h-3 w-3" aria-hidden="true" />
+          </a>
+        </div>
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,77 @@
+import React from "react";
+
+import { cn } from "@/lib/utils";
+
+export interface IssueSeveritySummaryFindings {
+  total: number;
+  critical: number;
+  high: number;
+  medium: number;
+  low: number;
+}
+
+interface IssueSeveritySummaryProps {
+  findings: IssueSeveritySummaryFindings;
+  className?: string;
+  /** Noun for the total count (e.g. "issues", "CVEs"). Defaults to "issues". */
+  unit?: string;
+  /** Optional content rendered at the end of the count row (e.g. a KEV badge). */
+  trailing?: React.ReactNode;
+}
+
+const SEVERITIES = [
+  { key: "critical", label: "critical", dotClass: "bg-red-500", textClass: "text-red-500" },
+  { key: "high", label: "high", dotClass: "bg-orange-500", textClass: "text-orange-500" },
+  { key: "medium", label: "medium", dotClass: "bg-yellow-500", textClass: "text-yellow-500" },
+  { key: "low", label: "low", dotClass: "bg-blue-500", textClass: "text-blue-500" },
+] as const;
+
+export function IssueSeveritySummary({
+  findings,
+  className,
+  unit = "issues",
+  trailing,
+}: IssueSeveritySummaryProps) {
+  if (findings.total <= 0) return null;
+
+  return (
+    <div className={cn("space-y-3", className)}>
+      <div className="flex flex-wrap items-center gap-x-8 gap-y-3">
+        <div className="flex items-center gap-2">
+          <span className="text-2xl font-semibold text-white tabular-nums">{findings.total}</span>
+          <span className="text-sm text-[#666]">{unit}</span>
+        </div>
+        <div className="flex flex-wrap items-center gap-x-6 gap-y-2">
+          {SEVERITIES.map(({ key, label, dotClass, textClass }) => {
+            const count = findings[key];
+            if (count <= 0) return null;
+
+            return (
+              <div key={key} className="flex items-center gap-1.5">
+                <div className={cn("w-2 h-2 rounded-full", dotClass)} aria-hidden="true" />
+                <span className={cn("text-sm tabular-nums", textClass)}>{count}</span>
+                <span className="text-xs text-[#555]">{label}</span>
+              </div>
+            );
+          })}
+        </div>
+        {trailing ? <div className="flex items-center gap-2">{trailing}</div> : null}
+      </div>
+
+      <div className="h-1.5 rounded-full bg-[#222] overflow-hidden flex">
+        {SEVERITIES.map(({ key, dotClass }) => {
+          const count = findings[key];
+          if (count <= 0) return null;
+
+          return (
+            <div
+              key={key}
+              className={cn("h-full", dotClass)}
+              style={{ width: `${(count / findings.total) * 100}%` }}
+            />
+          );
+        })}
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,176 @@
+import { useState } from "react";
+import { History, ChevronRight, Terminal } from "lucide-react";
+import type { RunListEntry, RunsPayload, RunSeverityCounts } from "@/data/serverSource";
+import { runTitle } from "@/lib/target-utils";
+import { trackCta } from "@/lib/cta";
+import EmailVerifyInline from "@/components/EmailVerifyInline";
+
+/**
+ * "Past runs" panel. Unverified users see a tease with the run count and a
+ * verify affordance (the launched run stays fully visible; the CLI
+ * `strix view <name>` still works). Verified users get the full history and can
+ * switch the active run, which threads ?run=<name> through the data fetches.
+ */
+
+const SEV = [
+  { key: "critical", dot: "bg-red-500", text: "text-red-500" },
+  { key: "high", dot: "bg-orange-500", text: "text-orange-500" },
+  { key: "medium", dot: "bg-yellow-500", text: "text-yellow-500" },
+  { key: "low", dot: "bg-blue-500", text: "text-blue-500" },
+] as const;
+
+function SeverityChips({ counts }: { counts: RunSeverityCounts }) {
+  const shown = SEV.filter((s) => counts[s.key] > 0);
+  if (shown.length === 0) {
+    return <span className="text-xs text-[#555]">No findings</span>;
+  }
+  return (
+    <div className="flex items-center gap-3">
+      {shown.map((s) => (
+        <div key={s.key} className="flex items-center gap-1.5">
+          <span className={`h-2 w-2 rounded-full ${s.dot}`} aria-hidden="true" />
+          <span className={`text-xs tabular-nums ${s.text}`}>{counts[s.key]}</span>
+        </div>
+      ))}
+    </div>
+  );
+}
+
+function formatDate(iso: string | null): string | null {
+  if (!iso) return null;
+  const normalized = iso.trim().replace(" UTC", "Z").replace(" ", "T");
+  const d = new Date(normalized);
+  if (Number.isNaN(d.getTime())) return null;
+  return d.toLocaleString(undefined, {
+    month: "short",
+    day: "numeric",
+    hour: "numeric",
+    minute: "2-digit",
+  });
+}
+
+/**
+ * Relative time ("just now" / "5m ago" / "3h ago" / "2d ago"), falling back to
+ * the absolute date for anything older than a week (mirrors the pro app).
+ */
+function formatTimeAgo(iso: string | null): string | null {
+  if (!iso) return null;
+  const normalized = iso.trim().replace(" UTC", "Z").replace(" ", "T");
+  const d = new Date(normalized);
+  if (Number.isNaN(d.getTime())) return null;
+  const diffMs = Date.now() - d.getTime();
+  const mins = Math.floor(diffMs / 60000);
+  if (mins < 1) return "just now";
+  if (mins < 60) return `${mins}m ago`;
+  const hours = Math.floor(mins / 60);
+  if (hours < 24) return `${hours}h ago`;
+  const days = Math.floor(hours / 24);
+  if (days < 7) return `${days}d ago`;
+  return formatDate(iso);
+}
+
+interface PastRunsViewProps {
+  runs: RunsPayload | null;
+  activeRun: string | null;
+  onSelectRun: (name: string) => void;
+  onVerified: () => void;
+}
+
+export default function PastRunsView({
+  runs,
+  activeRun,
+  onSelectRun,
+  onVerified,
+}: PastRunsViewProps) {
+  const count = runs?.count ?? 0;
+  const [showVerify, setShowVerify] = useState(false);
+
+  if (!runs || runs.locked) {
+    return (
+      <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-8 text-center">
+        <div
+          className="mx-auto mb-4 flex h-11 w-11 items-center justify-center rounded-xl"
+          style={{ border: "1px solid #2a2a2a", background: "rgba(255,255,255,0.04)" }}
+        >
+          <History className="h-5 w-5 text-[#888]" aria-hidden="true" />
+        </div>
+        <h2 className="text-base font-semibold text-white">Browse every run on this machine</h2>
+        <p className="mx-auto mt-1.5 max-w-md text-sm text-[#888]">
+          You have {count} past {count === 1 ? "run" : "runs"} on this machine.
+        </p>
+        {showVerify ? (
+          <>
+            <p className="mx-auto mt-3 max-w-sm text-xs text-[#666]">
+              Verify your email with a one-time code to unlock the full history.
+            </p>
+            <EmailVerifyInline onVerified={onVerified} />
+          </>
+        ) : (
+          <button
+            onClick={() => {
+              trackCta("history_unlock", "past_runs");
+              setShowVerify(true);
+            }}
+            className="mt-4 cursor-pointer rounded-lg bg-white px-4 py-2 text-sm font-semibold text-black transition-opacity hover:opacity-90"
+          >
+            View runs
+          </button>
+        )}
+        <p className="mt-4 flex items-center justify-center gap-1.5 text-xs text-[#555]">
+          <Terminal className="h-3.5 w-3.5" aria-hidden="true" />
+          Or open one from the CLI with{" "}
+          <code className="font-mono text-[#888]">strix view &lt;name&gt;</code>
+        </p>
+      </div>
+    );
+  }
+
+  if (runs.runs.length === 0) {
+    return (
+      <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-8 text-center text-sm text-[#888]">
+        No past runs found on this machine yet.
+      </div>
+    );
+  }
+
+  return (
+    <div className="space-y-2">
+      {runs.runs.map((run: RunListEntry) => {
+        const active = run.name === activeRun;
+        const date = formatTimeAgo(run.start_time) ?? formatTimeAgo(run.end_time);
+        const title = runTitle(run.target, run.name);
+        return (
+          <button
+            key={run.name}
+            onClick={() => onSelectRun(run.name)}
+            className={`group flex w-full cursor-pointer items-center gap-4 rounded-lg border px-4 py-3 text-left transition-colors ${
+              active
+                ? "border-[#444] bg-[rgba(255,255,255,0.04)]"
+                : "border-[#222] bg-[rgba(255,255,255,0.02)] hover:border-[#444]"
+            }`}
+          >
+            <div className="min-w-0 flex-1">
+              <div className="flex items-center gap-2">
+                <span className="truncate text-sm font-medium text-white">{title}</span>
+                {active && (
+                  <span className="rounded-full px-1.5 py-0.5 text-[10px] font-semibold uppercase tracking-wide text-emerald-400" style={{ border: "1px solid rgba(16,185,129,0.3)" }}>
+                    Active
+                  </span>
+                )}
+              </div>
+              <div className="mt-0.5 flex flex-wrap items-center gap-x-2 gap-y-0.5 text-xs text-[#666]">
+                {run.scan_mode && <span className="capitalize">{run.scan_mode}</span>}
+                {run.scan_mode && (date || run.status) && <span className="text-[#333]">·</span>}
+                {date && <span>{date}</span>}
+                {date && run.status && <span className="text-[#333]">·</span>}
+                {run.status && <span className="capitalize">{run.status}</span>}
+              </div>
+            </div>
+            <SeverityChips counts={run.severity_counts} />
+            <ChevronRight className="h-4 w-4 flex-shrink-0 text-[#555] transition-colors group-hover:text-[#aaa]" aria-hidden="true" />
+          </button>
+        );
+      })}
+    </div>
+  );
+}
@@ -0,0 +1,184 @@
+import React, { useState } from "react";
+import { ArrowUpRight } from "lucide-react";
+import { SIGNUP_URL, ctaUrl, trackCta } from "@/lib/cta";
+import type { ProFeature } from "@/lib/pro-features";
+
+/**
+ * Shared Pro CTA primitives. Every Pro item is a direct link-out to the cloud
+ * sign-up in a new tab with a hover tooltip one-liner (no modal, no lock icon).
+ * Built once here and reused by the sidebar Platform section, the top upsell
+ * row, and the inline CTAs in the tabs.
+ */
+
+/** Small tier pill ("Pro" or "Enterprise"). Deliberately not a padlock. */
+export function ProTag({ label = "Pro", className = "" }: { label?: string; className?: string }) {
+  return (
+    <span
+      className={`inline-flex items-center rounded-full px-1.5 py-0.5 text-[10px] font-semibold uppercase tracking-wide text-[#aaa] ${className}`}
+      style={{ border: "1px solid #2a2a2a", background: "rgba(255,255,255,0.04)" }}
+    >
+      {label}
+    </span>
+  );
+}
+
+/**
+ * Lightweight hover tooltip. Wraps a trigger and reveals `text` above it on
+ * hover/focus. Plain Tailwind + local state (no radix vendored).
+ */
+export function Tooltip({
+  text,
+  children,
+  className = "",
+}: {
+  text: string;
+  children: React.ReactNode;
+  className?: string;
+}) {
+  const [open, setOpen] = useState(false);
+  return (
+    <span
+      className={`relative inline-flex ${className}`}
+      onMouseEnter={() => setOpen(true)}
+      onMouseLeave={() => setOpen(false)}
+      onFocus={() => setOpen(true)}
+      onBlur={() => setOpen(false)}
+    >
+      {children}
+      {open && (
+        <span
+          role="tooltip"
+          className="pointer-events-none absolute bottom-full left-1/2 z-50 mb-2 w-max max-w-[240px] -translate-x-1/2 rounded-md px-2.5 py-1.5 text-xs text-[#ddd] shadow-lg"
+          style={{ border: "1px solid #2a2a2a", background: "#0a0a0a" }}
+        >
+          {text}
+        </span>
+      )}
+    </span>
+  );
+}
+
+export interface ProItem {
+  title: string;
+  desc: string;
+  slug: string;
+  icon: React.ElementType;
+}
+
+/**
+ * Card-style Pro feature tile: icon + name + one-liner + Pro tag + arrow.
+ * Used in the top upsell row and inline CTA grids.
+ */
+export function ProTile({ item, surface }: { item: ProItem; surface?: string }) {
+  const Icon = item.icon;
+  return (
+    <a
+      href={ctaUrl(SIGNUP_URL, item.slug)}
+      target="_blank"
+      rel="noopener noreferrer"
+      onClick={() => trackCta(item.slug, surface)}
+      title={item.desc}
+      className="group block cursor-pointer rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-4 text-left transition-colors hover:border-[#444]"
+    >
+      <div className="mb-2 flex items-center justify-between">
+        <Icon className="h-4 w-4 text-[#888] transition-colors group-hover:text-white" aria-hidden="true" />
+        <div className="flex items-center gap-1.5">
+          <ProTag />
+          <ArrowUpRight className="h-3.5 w-3.5 text-[#555] transition-colors group-hover:text-[#aaa]" aria-hidden="true" />
+        </div>
+      </div>
+      <p className="text-sm font-medium text-white">{item.title}</p>
+      <p className="mt-0.5 text-xs text-[#666]">{item.desc}</p>
+    </a>
+  );
+}
+
+/**
+ * Sidebar-row Pro item: a two-line row (icon + label + short one-liner
+ * underneath) with a small right-aligned tier tag. Opens the in-app
+ * FeatureDetail view via onClick (no link-out) so it sits uniformly beside the
+ * run/local rows in the themed nav list.
+ */
+export function ProNavItem({
+  feature,
+  active,
+  onClick,
+  collapsed = false,
+}: {
+  feature: ProFeature;
+  active?: boolean;
+  onClick: () => void;
+  collapsed?: boolean;
+}) {
+  const Icon = feature.icon;
+  if (collapsed) {
+    return (
+      <button
+        onClick={onClick}
+        title={`${feature.title} (${feature.tier})`}
+        className={`group flex w-full cursor-pointer items-center justify-center rounded-md px-2.5 py-2 transition-colors ${
+          active
+            ? "text-white"
+            : "text-[#888] hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+        }`}
+        style={active ? { background: "rgba(255,255,255,0.12)" } : undefined}
+      >
+        <Icon className="h-4 w-4 flex-shrink-0" aria-hidden="true" />
+      </button>
+    );
+  }
+  return (
+    <button
+      onClick={onClick}
+      className={`group flex w-full cursor-pointer items-start gap-2.5 rounded-md px-2.5 py-1.5 text-left transition-colors ${
+        active
+          ? "text-white"
+          : "text-[#888] hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+      }`}
+      style={active ? { background: "rgba(255,255,255,0.12)" } : undefined}
+    >
+      <Icon className="mt-0.5 h-4 w-4 flex-shrink-0" aria-hidden="true" />
+      <span className="min-w-0 flex-1">
+        <span className="flex items-center gap-1.5">
+          <span className="flex-1 truncate text-sm">{feature.title}</span>
+          <ProTag label={feature.tier} />
+        </span>
+        <span className="mt-0.5 block text-[11px] leading-snug text-[#666]">{feature.navDesc}</span>
+      </span>
+    </button>
+  );
+}
+
+/**
+ * Inline Pro CTA button (compact). Used in the finding detail and per-surface
+ * rows where a full card is too heavy.
+ */
+export function ProInlineCta({
+  label,
+  desc,
+  slug,
+  icon: Icon,
+  surface,
+}: {
+  label: string;
+  desc: string;
+  slug: string;
+  icon: React.ElementType;
+  surface?: string;
+}) {
+  return (
+    <Tooltip text={desc}>
+      <a
+        href={ctaUrl(SIGNUP_URL, slug)}
+        target="_blank"
+        rel="noopener noreferrer"
+        onClick={() => trackCta(slug, surface)}
+        className="group inline-flex items-center gap-2 rounded-lg border border-[#222] bg-[rgba(255,255,255,0.02)] px-3 py-2 text-sm text-[#aaa] transition-colors hover:border-[#444] hover:text-white"
+      >
+        <Icon className="h-4 w-4 text-[#888] transition-colors group-hover:text-white" aria-hidden="true" />
+        <span>{label}</span>
+        <ProTag className="ml-0.5" />
+      </a>
+    </Tooltip>
+  );
+}
@@ -0,0 +1,185 @@
+import { Info } from "lucide-react";
+import { formatNumber } from "@/lib/display-number";
+
+/**
+ * "Run details" card for the Overview tab: the launch configuration the run was
+ * started with (targets, instruction, scope, mode) and its LLM usage + cost.
+ * Everything is read defensively from the raw run.json record, which may be
+ * partial while a scan is still live.
+ */
+
+type Rec = Record<string, unknown>;
+
+function rec(v: unknown): Rec {
+  return v && typeof v === "object" && !Array.isArray(v) ? (v as Rec) : {};
+}
+function arr(v: unknown): unknown[] {
+  return Array.isArray(v) ? v : [];
+}
+function str(v: unknown): string | null {
+  return typeof v === "string" && v.trim() ? v : null;
+}
+function num(v: unknown): number | null {
+  return typeof v === "number" && Number.isFinite(v) ? v : null;
+}
+function humanize(s: string): string {
+  return s.replace(/_/g, " ");
+}
+function cap(s: string | null): string | null {
+  return s ? s.charAt(0).toUpperCase() + s.slice(1) : s;
+}
+function fmtDuration(seconds: number | null): string {
+  if (seconds == null || seconds < 0) return "n/a";
+  const h = Math.floor(seconds / 3600);
+  const m = Math.floor((seconds % 3600) / 60);
+  const s = seconds % 60;
+  if (h) return `${h}h ${m}m ${s}s`;
+  if (m) return `${m}m ${s}s`;
+  return `${s}s`;
+}
+
+function Field({ label, children }: { label: string; children: React.ReactNode }) {
+  return (
+    <div className="grid grid-cols-[7rem_1fr] gap-3 items-baseline">
+      <dt className="text-[11px] uppercase tracking-wide text-[#666]">{label}</dt>
+      <dd className="min-w-0 break-words text-sm text-[#ddd]">{children}</dd>
+    </div>
+  );
+}
+
+export function RunDetails({
+  raw,
+  durationSeconds,
+}: {
+  raw: Rec;
+  durationSeconds: number | null;
+}) {
+  // Configuration (launch inputs)
+  const targets = arr(raw.targets_info).map((t) => {
+    const o = rec(t);
+    const display = str(o.original) ?? str(rec(o.details).target_url) ?? "unknown target";
+    const type = str(o.type);
+    return { display, type: type ? humanize(type) : null };
+  });
+  const instruction = str(raw.instruction);
+  const scanMode = cap(str(raw.scan_mode));
+  const scopeMode = str(raw.scope_mode);
+  const diff = rec(raw.diff_scope);
+  const diffActive = diff.active === true;
+  const diffMode = str(diff.mode);
+  const diffBase = str(raw.diff_base);
+  const nonInteractive = raw.non_interactive === true;
+  const localSources = arr(raw.local_sources).map((x) => String(x)).filter(Boolean);
+  const status = cap(str(raw.status));
+
+  let scope = scopeMode ?? "auto";
+  if (diffActive) {
+    scope += ` (diff${diffMode ? `: ${diffMode}` : ""}${diffBase ? ` vs ${diffBase}` : ""})`;
+  }
+
+  // Usage & cost
+  const usage = rec(raw.llm_usage);
+  const hasUsage = Object.keys(usage).length > 0;
+  const agents = arr(usage.agents).map(rec);
+  const models = Array.from(
+    new Set(agents.map((a) => str(a.model)).filter((m): m is string => !!m))
+  );
+  const requests = num(usage.requests);
+  const inputTokens = num(usage.input_tokens);
+  const cached = num(rec(arr(usage.input_tokens_details)[0]).cached_tokens);
+  const outputTokens = num(usage.output_tokens);
+  const reasoning = num(rec(arr(usage.output_tokens_details)[0]).reasoning_tokens);
+  const totalTokens = num(usage.total_tokens);
+  const cost = num(usage.cost);
+
+  const sub = (n: number, word: string) => (
+    <span className="text-[#666]"> ({formatNumber(n)} {word})</span>
+  );
+
+  return (
+    <div className="rounded-xl border border-[#222] bg-[rgba(255,255,255,0.02)] p-5">
+      <div className="mb-4 flex items-center gap-2">
+        <Info className="h-4 w-4 text-[#888]" aria-hidden="true" />
+        <h2 className="text-sm font-semibold text-white">Run details</h2>
+      </div>
+
+      <div className="grid grid-cols-1 gap-x-8 gap-y-6 md:grid-cols-2">
+        <section>
+          <h3 className="mb-3 text-[11px] font-semibold uppercase tracking-wide text-[#555]">
+            Configuration
+          </h3>
+          <dl className="space-y-2.5">
+            {targets.length > 0 && (
+              <Field label="Targets">
+                <div className="space-y-1">
+                  {targets.map((t, i) => (
+                    <div key={i} className="flex flex-wrap items-center gap-2">
+                      <span className="font-mono text-[#ddd]">{t.display}</span>
+                      {t.type && (
+                        <span className="rounded-full border border-[#2a2a2a] px-1.5 py-0.5 text-[10px] text-[#888]">
+                          {t.type}
+                        </span>
+                      )}
+                    </div>
+                  ))}
+                </div>
+              </Field>
+            )}
+            <Field label="Instruction">
+              {instruction ? (
+                <span className="whitespace-pre-wrap">{instruction}</span>
+              ) : (
+                <span className="text-[#666]">None</span>
+              )}
+            </Field>
+            {scanMode && <Field label="Scan mode">{scanMode}</Field>}
+            <Field label="Scope">{scope}</Field>
+            <Field label="Mode">{nonInteractive ? "Non-interactive" : "Interactive"}</Field>
+            {localSources.length > 0 && (
+              <Field label="Local sources">
+                <div className="space-y-0.5 font-mono text-[#ddd]">
+                  {localSources.map((s, i) => (
+                    <div key={i}>{s}</div>
+                  ))}
+                </div>
+              </Field>
+            )}
+            {status && <Field label="Status">{status}</Field>}
+          </dl>
+        </section>
+
+        <section>
+          <h3 className="mb-3 text-[11px] font-semibold uppercase tracking-wide text-[#555]">
+            Usage &amp; cost
+          </h3>
+          {hasUsage ? (
+            <dl className="space-y-2.5 tabular-nums">
+              <Field label="Model">{models.length ? models.join(", ") : "n/a"}</Field>
+              <Field label="Run time">{fmtDuration(durationSeconds)}</Field>
+              {requests != null && <Field label="Requests">{formatNumber(requests)}</Field>}
+              {inputTokens != null && (
+                <Field label="Input tokens">
+                  {formatNumber(inputTokens)}
+                  {cached != null && sub(cached, "cached")}
+                </Field>
+              )}
+              {outputTokens != null && (
+                <Field label="Output tokens">
+                  {formatNumber(outputTokens)}
+                  {reasoning != null && sub(reasoning, "reasoning")}
+                </Field>
+              )}
+              {totalTokens != null && <Field label="Total tokens">{formatNumber(totalTokens)}</Field>}
+              {cost != null && <Field label="Cost">${cost.toFixed(2)}</Field>}
+              {agents.length > 0 && <Field label="Agents">{formatNumber(agents.length)}</Field>}
+            </dl>
+          ) : (
+            <p className="text-sm text-[#666]">Not available yet.</p>
+          )}
+        </section>
+      </div>
+    </div>
+  );
+}
+
+export default RunDetails;
@@ -0,0 +1,283 @@
+import React, { useEffect, useState } from "react";
+import {
+  FileText,
+  Bug,
+  Waypoints,
+  History,
+  Mail,
+  ArrowUpRight,
+  LogOut,
+  ShieldCheck,
+  PanelLeftClose,
+  PanelLeft,
+} from "lucide-react";
+import { SIGNUP_URL, ctaUrl, trackCta } from "@/lib/cta";
+import { ProNavItem } from "@/components/ProCta";
+import { FEATURES, PLATFORM_ORDER } from "@/lib/pro-features";
+import type { View } from "@/App";
+
+/**
+ * Persistent left rail. A single, ungrouped, ordered list of uniform two-line
+ * rows (icon + label + short one-liner): the current run's views, the local
+ * run-history + email-report actions, then the platform features. No section
+ * headers. Tier is shown only by the inline Pro/Enterprise tag on platform
+ * rows. Matches App.tsx's dark palette.
+ *
+ * Can collapse to a narrow icon-only rail; the collapsed state persists in
+ * localStorage and each icon row keeps a `title` tooltip so the labels stay
+ * discoverable.
+ */
+
+const COLLAPSE_KEY = "strix_viewer_sidebar_collapsed";
+
+interface SidebarProps {
+  view: View;
+  onSelectView: (view: View) => void;
+  activeFeature: string | null;
+  onSelectFeature: (slug: string) => void;
+  issuesCount: number;
+  agentCount: number;
+  runCount: number;
+  verified: boolean;
+  email: string | null;
+  onOpenEmail: () => void;
+  onOpenHistory: () => void;
+  onForget: () => void;
+}
+
+export default function Sidebar({
+  view,
+  onSelectView,
+  activeFeature,
+  onSelectFeature,
+  issuesCount,
+  agentCount,
+  runCount,
+  verified,
+  email,
+  onOpenEmail,
+  onOpenHistory,
+  onForget,
+}: SidebarProps) {
+  const [collapsed, setCollapsed] = useState(false);
+
+  useEffect(() => {
+    try {
+      setCollapsed(localStorage.getItem(COLLAPSE_KEY) === "1");
+    } catch {
+      /* localStorage may be unavailable; default to expanded */
+    }
+  }, []);
+
+  const toggleCollapsed = () => {
+    setCollapsed((prev) => {
+      const next = !prev;
+      try {
+        localStorage.setItem(COLLAPSE_KEY, next ? "1" : "0");
+      } catch {
+        /* best-effort persistence */
+      }
+      return next;
+    });
+  };
+
+  return (
+    <aside
+      className={`hidden flex-shrink-0 border-r border-[#222] lg:block ${collapsed ? "w-14" : "w-72"}`}
+    >
+      <div className="sticky top-0 flex h-screen flex-col overflow-y-auto px-3 py-4">
+        {/* Header: wordmark + Explore full platform + signed-in chip */}
+        <div className="px-1.5">
+          <div className={`flex items-center ${collapsed ? "flex-col gap-2" : "justify-between"}`}>
+            <a
+              href={ctaUrl("https://app.strix.ai", "logo")}
+              target="_blank"
+              rel="noopener noreferrer"
+              onClick={() => trackCta("logo", "sidebar")}
+              className="flex items-center gap-1.5 opacity-90 transition-opacity hover:opacity-100"
+              title="Open Strix Cloud"
+            >
+              <img src="./logo.png" alt="Strix" className="h-8 w-10 object-cover" />
+              {!collapsed && (
+                <span className="text-base font-medium tracking-tight text-white">Strix</span>
+              )}
+            </a>
+            <button
+              onClick={toggleCollapsed}
+              title={collapsed ? "Expand sidebar" : "Collapse sidebar"}
+              aria-label={collapsed ? "Expand sidebar" : "Collapse sidebar"}
+              className="flex-shrink-0 cursor-pointer rounded-md p-1.5 text-[#666] transition-colors hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+            >
+              {collapsed ? (
+                <PanelLeft className="h-4 w-4" aria-hidden="true" />
+              ) : (
+                <PanelLeftClose className="h-4 w-4" aria-hidden="true" />
+              )}
+            </button>
+          </div>
+          <a
+            href={ctaUrl(SIGNUP_URL, "sidebar_start_free")}
+            target="_blank"
+            rel="noopener noreferrer"
+            onClick={() => trackCta("sidebar_start_free", "sidebar")}
+            title="Explore full platform"
+            className={`mt-3 flex w-full cursor-pointer items-center justify-center gap-1.5 rounded-lg bg-white font-semibold text-black transition-opacity hover:opacity-90 ${
+              collapsed ? "px-0 py-2" : "px-3 py-2 text-sm"
+            }`}
+          >
+            {!collapsed && "Explore full platform"}
+            <ArrowUpRight className="h-3.5 w-3.5" aria-hidden="true" />
+          </a>
+          {verified && email && (
+            collapsed ? (
+              <div
+                className="mt-2.5 flex items-center justify-center rounded-lg py-2"
+                style={{ border: "1px solid #222", background: "rgba(255,255,255,0.02)" }}
+                title={`Linked email: ${email}`}
+              >
+                <ShieldCheck className="h-3.5 w-3.5 flex-shrink-0 text-emerald-400" aria-hidden="true" />
+              </div>
+            ) : (
+              <div
+                className="mt-2.5 flex items-center gap-2 rounded-lg px-2.5 py-2"
+                style={{ border: "1px solid #222", background: "rgba(255,255,255,0.02)" }}
+              >
+                <ShieldCheck className="h-3.5 w-3.5 flex-shrink-0 text-emerald-400" aria-hidden="true" />
+                <div className="min-w-0 flex-1">
+                  <p className="truncate text-[11px] text-[#666]">Linked email</p>
+                  <p className="truncate text-xs text-[#aaa]" title={email}>{email}</p>
+                </div>
+                <button
+                  onClick={onForget}
+                  title="Forget this email on this machine"
+                  className="flex-shrink-0 cursor-pointer text-[#666] transition-colors hover:text-white"
+                  aria-label="Forget"
+                >
+                  <LogOut className="h-3.5 w-3.5" />
+                </button>
+              </div>
+            )
+          )}
+        </div>
+
+        {/* One single ordered list, no section headers. */}
+        <div className="mt-6 space-y-0.5">
+          <NavItem
+            icon={FileText}
+            label="Overview"
+            desc="This run's executive report"
+            active={view === "overview"}
+            onClick={() => onSelectView("overview")}
+            collapsed={collapsed}
+          />
+          <NavItem
+            icon={Bug}
+            label="Issues"
+            desc="Findings from this run"
+            count={issuesCount > 0 ? issuesCount : undefined}
+            active={view === "issues"}
+            onClick={() => onSelectView("issues")}
+            collapsed={collapsed}
+          />
+          {agentCount > 0 && (
+            <NavItem
+              icon={Waypoints}
+              label="Agents"
+              desc="What each agent did"
+              count={agentCount}
+              active={view === "agents"}
+              onClick={() => onSelectView("agents")}
+              collapsed={collapsed}
+            />
+          )}
+          <NavItem
+            icon={History}
+            label="Past runs"
+            desc="Every run on this machine"
+            count={runCount > 0 ? runCount : undefined}
+            active={view === "history"}
+            onClick={onOpenHistory}
+            collapsed={collapsed}
+          />
+          <NavItem
+            icon={Mail}
+            label="Email report"
+            desc="Get an encrypted PDF by email"
+            active={view === "email"}
+            onClick={onOpenEmail}
+            collapsed={collapsed}
+          />
+
+          {PLATFORM_ORDER.map((slug) => {
+            const feature = FEATURES[slug];
+            if (!feature) return null;
+            return (
+              <ProNavItem
+                key={slug}
+                feature={feature}
+                active={view === "feature" && activeFeature === slug}
+                onClick={() => onSelectFeature(slug)}
+                collapsed={collapsed}
+              />
+            );
+          })}
+        </div>
+      </div>
+    </aside>
+  );
+}
+
+function NavItem({
+  icon: Icon,
+  label,
+  desc,
+  count,
+  active,
+  onClick,
+  collapsed = false,
+}: {
+  icon: React.ElementType;
+  label: string;
+  desc: string;
+  count?: number;
+  active?: boolean;
+  onClick: () => void;
+  collapsed?: boolean;
+}) {
+  if (collapsed) {
+    return (
+      <button
+        onClick={onClick}
+        title={count != null ? `${label} (${count})` : label}
+        className={`flex w-full cursor-pointer items-center justify-center rounded-md px-2.5 py-2 transition-colors ${
+          active
+            ? "text-white"
+            : "text-[#888] hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+        }`}
+        style={active ? { background: "rgba(255,255,255,0.12)" } : undefined}
+      >
+        <Icon className="h-4 w-4 flex-shrink-0" aria-hidden="true" />
+      </button>
+    );
+  }
+  return (
+    <button
+      onClick={onClick}
+      className={`flex w-full cursor-pointer items-start gap-2.5 rounded-md px-2.5 py-1.5 text-left transition-colors ${
+        active
+          ? "text-white"
+          : "text-[#888] hover:bg-[rgba(255,255,255,0.06)] hover:text-white"
+      }`}
+      style={active ? { background: "rgba(255,255,255,0.12)" } : undefined}
+    >
+      <Icon className="mt-0.5 h-4 w-4 flex-shrink-0" aria-hidden="true" />
+      <span className="min-w-0 flex-1">
+        <span className="flex items-center gap-1.5">
+          <span className="flex-1 truncate text-sm">{label}</span>
+          {count != null && <span className="text-xs text-[#666] tabular-nums">{count}</span>}
+        </span>
+        <span className="mt-0.5 block text-[11px] leading-snug text-[#666]">{desc}</span>
+      </span>
+    </button>
+  );
+}
@@ -0,0 +1,51 @@
+import { useState } from "react";
+import { ShieldCheck, X } from "lucide-react";
+
+const DISMISS_KEY = "strix_viewer_trust_dismissed";
+
+/**
+ * One-time privacy notice, shown as a toast pinned over the sidebar. Dismissing
+ * it persists to localStorage so it never returns on reload or view changes.
+ */
+export function TrustToast({ message }: { message: string }) {
+  const [dismissed, setDismissed] = useState<boolean>(() => {
+    try {
+      return localStorage.getItem(DISMISS_KEY) === "1";
+    } catch {
+      return false;
+    }
+  });
+
+  if (dismissed) return null;
+
+  const dismiss = () => {
+    try {
+      localStorage.setItem(DISMISS_KEY, "1");
+    } catch {
+      /* non-fatal: worst case the toast shows again next session */
+    }
+    setDismissed(true);
+  };
+
+  return (
+    <div
+      className="fixed bottom-3 left-3 z-[60] max-w-xs rounded-lg bg-[#0a0a0a] p-3 shadow-2xl"
+      style={{ border: "1px solid #2a2a2a" }}
+      role="status"
+    >
+      <div className="flex gap-2.5">
+        <ShieldCheck className="mt-0.5 h-4 w-4 flex-shrink-0 text-emerald-400" aria-hidden="true" />
+        <p className="text-xs leading-relaxed text-[#aaa]">{message}</p>
+        <button
+          onClick={dismiss}
+          aria-label="Dismiss"
+          className="-mr-0.5 -mt-0.5 flex-shrink-0 cursor-pointer rounded p-0.5 text-[#666] transition-colors hover:text-white"
+        >
+          <X className="h-3.5 w-3.5" />
+        </button>
+      </div>
+    </div>
+  );
+}
+
+export default TrustToast;
@@ -0,0 +1,117 @@
+import { useCallback, useEffect, useRef } from "react";
+import { X } from "lucide-react";
+import { AgentTranscript } from "./AgentTranscript";
+import { ScanPromptComposer } from "./ScanPromptComposer";
+import type { TranscriptAgent, TranscriptEvent } from "@/data/serverSource";
+
+/** Status -> the small leading dot color, matching the graph node styling. */
+const STATUS_DOT: Record<string, string> = {
+  completed: "bg-emerald-400",
+  running: "bg-blue-400",
+  waiting: "bg-yellow-400",
+  stopped: "bg-[#888]",
+  crashed: "bg-red-400",
+  failed: "bg-red-400",
+};
+
+/** Consider the user "at the bottom" within this many px. */
+const NEAR_BOTTOM_PX = 80;
+
+/**
+ * Overlay modal showing a single agent's full transcript. Matches the cloud
+ * app: a fixed-size panel with a pinned header (status dot + agent name), the
+ * transcript scrolling beneath it, and a footer. Auto-scrolls to follow new
+ * activity while the user is near the bottom (so a live run trails). Closes on
+ * backdrop click, the X button, or Escape.
+ */
+export function AgentDetailModal({
+  agent,
+  events,
+  steerable,
+  onClose,
+}: {
+  agent: TranscriptAgent;
+  events: TranscriptEvent[];
+  steerable: boolean;
+  onClose: () => void;
+}) {
+  const scrollRef = useRef<HTMLDivElement>(null);
+  const nearBottom = useRef(false);
+
+  const handleScroll = useCallback(() => {
+    const el = scrollRef.current;
+    if (!el) return;
+    nearBottom.current = el.scrollHeight - el.scrollTop - el.clientHeight < NEAR_BOTTOM_PX;
+  }, []);
+
+  // Follow new activity when the user is near the bottom (live trailing).
+  useEffect(() => {
+    const el = scrollRef.current;
+    if (!el || !nearBottom.current) return;
+    requestAnimationFrame(() => {
+      el.scrollTo({ top: el.scrollHeight, behavior: "smooth" });
+    });
+  }, [events]);
+
+  useEffect(() => {
+    const onKey = (e: KeyboardEvent) => {
+      if (e.key === "Escape") onClose();
+    };
+    document.addEventListener("keydown", onKey);
+    const prevOverflow = document.body.style.overflow;
+    document.body.style.overflow = "hidden";
+    return () => {
+      document.removeEventListener("keydown", onKey);
+      document.body.style.overflow = prevOverflow;
+    };
+  }, [onClose]);
+
+  return (
+    <div
+      className="fixed inset-0 z-50 flex items-center justify-center bg-black/70 p-4 sm:p-8"
+      onClick={onClose}
+      role="dialog"
+      aria-modal="true"
+      aria-label={`Agent ${agent.name}`}
+    >
+      <div
+        className="relative flex h-[80vh] w-full max-w-5xl flex-col rounded-xl border border-[#222] bg-[#0a0a0a] shadow-2xl"
+        onClick={(e) => e.stopPropagation()}
+      >
+        <div className="flex items-center justify-between gap-3 border-b border-[#222] px-5 py-3.5">
+          <div className="flex min-w-0 items-center gap-2">
+            <span
+              className={`h-2 w-2 flex-shrink-0 rounded-full ${STATUS_DOT[agent.status] ?? "bg-[#888]"}`}
+            />
+            <span className="truncate text-sm font-semibold text-white">{agent.name}</span>
+            <span className="flex-shrink-0 font-mono text-xs text-[#555]">{agent.id}</span>
+          </div>
+          <button
+            type="button"
+            onClick={onClose}
+            aria-label="Close"
+            className="flex-shrink-0 rounded-md p-1 text-[#888] transition-colors hover:bg-[#1a1a1a] hover:text-white"
+          >
+            <X className="h-4 w-4" />
+          </button>
+        </div>
+
+        <div ref={scrollRef} onScroll={handleScroll} className="flex-1 overflow-y-auto p-5">
+          <AgentTranscript agent={agent} events={events} showHeader={false} />
+        </div>
+
+        {steerable && (
+          <div className="border-t border-[#222] px-5 py-3">
+            <ScanPromptComposer
+              agents={[agent]}
+              fixedAgentId={agent.id}
+              className="mt-0"
+            />
+          </div>
+        )}
+      </div>
+    </div>
+  );
+}
+
+export default AgentDetailModal;
@@ -0,0 +1,254 @@
+"use client";
+
+import { useCallback, useEffect, useRef, useState } from "react";
+import {
+  ReactFlow,
+  Background,
+  Controls,
+  MiniMap,
+  useNodesState,
+  useEdgesState,
+  useReactFlow,
+  type Node,
+  type Edge,
+} from "@xyflow/react";
+import dagre from "@dagrejs/dagre";
+import AgentNodeComponent from "./AgentNode";
+import GraphSkeleton from "./GraphSkeleton";
+import type { AgentNode } from "@/types/events";
+
+import "@xyflow/react/dist/style.css";
+
+const NODE_WIDTH = 260;
+const NODE_HEIGHT = 80;
+
+const nodeTypes = { agentNode: AgentNodeComponent };
+
+function getLayoutedElements(
+  agents: Map<string, AgentNode>,
+  selectedAgentId: string | null
+) {
+  const g = new dagre.graphlib.Graph();
+  g.setDefaultEdgeLabel(() => ({}));
+  g.setGraph({ rankdir: "TB", nodesep: 60, ranksep: 80 });
+
+  const nodes: Node[] = [];
+  const edges: Edge[] = [];
+
+  for (const [id, agent] of agents) {
+    g.setNode(id, { width: NODE_WIDTH, height: NODE_HEIGHT });
+    nodes.push({
+      id,
+      type: "agentNode",
+      position: { x: 0, y: 0 },
+      data: { ...agent, isSelected: id === selectedAgentId },
+    });
+
+    if (agent.parentId && agents.has(agent.parentId)) {
+      const edgeId = `${agent.parentId}->${id}`;
+      g.setEdge(agent.parentId, id);
+      edges.push({
+        id: edgeId,
+        source: agent.parentId,
+        target: id,
+        style: { stroke: "#2a2a2a", strokeWidth: 1.5 },
+      });
+    }
+  }
+
+  dagre.layout(g);
+
+  for (const node of nodes) {
+    const pos = g.node(node.id);
+    if (pos) {
+      node.position = {
+        x: pos.x - NODE_WIDTH / 2,
+        y: pos.y - NODE_HEIGHT / 2,
+      };
+    }
+  }
+
+  return { nodes, edges };
+}
+
+const ZOOM_DURATION = 300;
+
+
+/** Centers viewport on the root node (no parentId) at a fixed zoom — only once on first load */
+function CenterOnRoot({ nodes }: { nodes: Node[] }) {
+  const { setCenter } = useReactFlow();
+  const hasCentered = useRef(false);
+  useEffect(() => {
+    if (nodes.length > 0 && !hasCentered.current) {
+      const root = nodes.find((n) => !(n.data as Record<string, unknown>).parentId);
+      const target = root ?? nodes[0];
+      hasCentered.current = true;
+      const cx = target.position.x + NODE_WIDTH / 2;
+      const cy = target.position.y + NODE_HEIGHT / 2;
+      setTimeout(() => setCenter(cx, cy, { zoom: 0.85, duration: 400 }), 60);
+    }
+  }, [nodes, setCenter]);
+  return null;
+}
+
+function SmoothControls() {
+  const { zoomIn, zoomOut, fitView } = useReactFlow();
+  return (
+    <Controls
+      position="bottom-right"
+      showZoom={false}
+      showFitView={false}
+      showInteractive={false}
+      className="!bg-transparent !border-none !shadow-none"
+    >
+      <div className="flex flex-col overflow-hidden rounded-lg border border-[#222]">
+        <button onClick={() => zoomIn({ duration: ZOOM_DURATION })} className="flex items-center justify-center w-7 h-7 bg-[#111] text-white hover:bg-[#2a2a2a] transition-colors" title="Zoom in">
+          <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" strokeWidth={2} className="w-3.5 h-3.5"><path d="M12 5v14M5 12h14" /></svg>
+        </button>
+        <button onClick={() => zoomOut({ duration: ZOOM_DURATION })} className="flex items-center justify-center w-7 h-7 bg-[#111] text-white hover:bg-[#2a2a2a] border-y border-[#222] transition-colors" title="Zoom out">
+          <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" strokeWidth={2} className="w-3.5 h-3.5"><path d="M5 12h14" /></svg>
+        </button>
+        <button onClick={() => fitView({ padding: 0.3, duration: ZOOM_DURATION })} className="flex items-center justify-center w-7 h-7 bg-[#111] text-white hover:bg-[#2a2a2a] transition-colors" title="Fit view">
+          <svg viewBox="0 0 24 24" fill="none" stroke="currentColor" strokeWidth={2} className="w-3.5 h-3.5"><path d="M15 3h6v6M9 21H3v-6M21 3l-7 7M3 21l7-7" /></svg>
+        </button>
+      </div>
+    </Controls>
+  );
+}
+
+interface AgentGraphProps {
+  agents: Map<string, AgentNode>;
+  selectedAgentId: string | null;
+  onSelectAgent: (id: string | null) => void;
+  eventsLoaded?: boolean;
+  eventsEmpty?: boolean;
+  scanCompleted?: boolean;
+}
+
+export default function AgentGraph({
+  agents,
+  selectedAgentId,
+  onSelectAgent,
+  eventsLoaded,
+  eventsEmpty,
+  scanCompleted,
+}: AgentGraphProps) {
+  const [nodes, setNodes, onNodesChange] = useNodesState<Node>([]);
+  const [edges, setEdges, onEdgesChange] = useEdgesState<Edge>([]);
+
+  useEffect(() => {
+    if (agents.size === 0) return;
+    const { nodes: ln, edges: le } = getLayoutedElements(agents, selectedAgentId);
+    setNodes(ln);
+    setEdges(le);
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [agents.size, setNodes, setEdges]);
+
+  // Sync agent data (status, name, etc.) into existing nodes without re-layout
+  useEffect(() => {
+    if (agents.size === 0) return;
+    setNodes((nds) =>
+      nds.map((n) => {
+        const agent = agents.get(n.id);
+        if (!agent) return n;
+        return { ...n, data: { ...agent, isSelected: n.id === selectedAgentId } };
+      })
+    );
+  }, [agents, selectedAgentId, setNodes]);
+
+  const nodeClickedRef = useRef(false);
+
+  const onNodeClick = useCallback(
+    (_: React.MouseEvent, node: Node) => {
+      nodeClickedRef.current = true;
+      onSelectAgent(node.id);
+    },
+    [onSelectAgent]
+  );
+
+  const onPaneClick = useCallback(() => {
+    if (nodeClickedRef.current) {
+      nodeClickedRef.current = false;
+      return;
+    }
+    onSelectAgent(null);
+  }, [onSelectAgent]);
+
+  // Convex responded, zero events — show empty state (not skeleton)
+  if (agents.size === 0 && eventsLoaded && eventsEmpty) {
+    return (
+      <div className="flex flex-col items-center justify-center h-full text-center px-4">
+        <div className="w-10 h-10 mb-3 rounded-full bg-[#111] flex items-center justify-center">
+          {scanCompleted ? (
+            <svg className="w-5 h-5 text-[#444]" fill="none" viewBox="0 0 24 24" stroke="currentColor" strokeWidth={1.5}>
+              <path strokeLinecap="round" strokeLinejoin="round" d="M3.75 6A2.25 2.25 0 0 1 6 3.75h2.25A2.25 2.25 0 0 1 10.5 6v2.25a2.25 2.25 0 0 1-2.25 2.25H6a2.25 2.25 0 0 1-2.25-2.25V6ZM3.75 15.75A2.25 2.25 0 0 1 6 13.5h2.25a2.25 2.25 0 0 1 2.25 2.25V18a2.25 2.25 0 0 1-2.25 2.25H6A2.25 2.25 0 0 1 3.75 18v-2.25ZM13.5 6a2.25 2.25 0 0 1 2.25-2.25H18A2.25 2.25 0 0 1 20.25 6v2.25A2.25 2.25 0 0 1 18 10.5h-2.25a2.25 2.25 0 0 1-2.25-2.25V6ZM13.5 15.75a2.25 2.25 0 0 1 2.25-2.25H18a2.25 2.25 0 0 1 2.25 2.25V18A2.25 2.25 0 0 1 18 20.25h-2.25a2.25 2.25 0 0 1-2.25-2.25v-2.25Z" />
+            </svg>
+          ) : (
+            <div className="w-2 h-2 rounded-full bg-blue-500 animate-pulse" />
+          )}
+        </div>
+        <p className="text-sm text-[#555]">
+          {scanCompleted
+            ? "Agent trace data is not available for this pentest"
+            : "Waiting for agent data\u2026"}
+        </p>
+      </div>
+    );
+  }
+
+  const showGraph = agents.size > 0;
+
+  return (
+    <div className="relative h-full">
+      {/* Skeleton overlay — fades out when graph is ready */}
+      <div
+        className={`absolute inset-0 z-10 transition-opacity duration-500 ${
+          showGraph ? "opacity-0 pointer-events-none" : "opacity-100"
+        }`}
+      >
+        <GraphSkeleton />
+      </div>
+
+      {/* Graph — fades in */}
+      <div
+        className={`h-full transition-opacity duration-500 ${
+          showGraph ? "opacity-100" : "opacity-0"
+        }`}
+      >
+    <ReactFlow
+      nodes={nodes}
+      edges={edges}
+      onNodesChange={onNodesChange}
+      onEdgesChange={onEdgesChange}
+      onNodeClick={onNodeClick}
+      onPaneClick={onPaneClick}
+      nodeTypes={nodeTypes}
+      nodesConnectable={false}
+      edgesFocusable={false}
+      edgesReconnectable={false}
+      minZoom={0.15}
+      maxZoom={1.5}
+      proOptions={{ hideAttribution: true }}
+      className="bg-black"
+    >
+      <Background color="#111" gap={20} />
+      <CenterOnRoot nodes={nodes} />
+      <SmoothControls />
+      <MiniMap
+        position="bottom-left"
+        nodeColor={(n) => {
+          const status = (n.data as Record<string, unknown>)?.status as string;
+          if (status === "running") return "#3b82f6";
+          if (status === "completed") return "#10b981";
+          if (status === "failed" || status === "error") return "#ef4444";
+          return "#555";
+        }}
+        maskColor="rgba(0,0,0,0.8)"
+        style={{ width: 80, height: 50 }}
+        className="!bg-[#0a0a0a] !border-[#222]"
+      />
+    </ReactFlow>
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,48 @@
+"use client";
+
+import { memo } from "react";
+import { Handle, Position, type NodeProps } from "@xyflow/react";
+import type { AgentNode as AgentNodeData } from "@/types/events";
+
+const STATUS_STYLES: Record<string, string> = {
+  running: "bg-blue-500",
+  completed: "bg-emerald-500",
+  failed: "bg-red-500",
+  error: "bg-red-500",
+};
+
+function AgentNodeComponent({ data, selected }: NodeProps) {
+  const agent = data as unknown as AgentNodeData & { isSelected: boolean };
+
+  return (
+    <div
+      className={`w-[260px] rounded-lg border px-4 py-3 transition-colors ${
+        agent.isSelected || selected
+          ? "border-white/30 bg-[#0a0a0a]"
+          : "border-[#222] bg-black hover:border-[#333]"
+      }`}
+    >
+      <Handle type="target" position={Position.Top} isConnectable={false} className={`!w-1.5 !h-1.5 !border-0 ${agent.parentId ? "!bg-[#444]" : "!bg-transparent"}`} />
+
+      <div className="flex items-center gap-2">
+        <span className="relative flex h-2 w-2 shrink-0">
+          <span
+            className={`absolute inline-flex h-full w-full rounded-full opacity-75 ${STATUS_STYLES[agent.status] ?? "bg-gray-500"} ${
+              agent.status === "running" ? "animate-ping" : ""
+            }`}
+          />
+          <span
+            className={`relative inline-flex h-2 w-2 rounded-full ${STATUS_STYLES[agent.status] ?? "bg-gray-500"}`}
+          />
+        </span>
+        <span className="text-sm font-semibold text-white leading-snug line-clamp-3">
+          {agent.name}
+        </span>
+      </div>
+
+      <Handle type="source" position={Position.Bottom} isConnectable={false} className={`!w-1.5 !h-1.5 !border-0 ${agent.children && agent.children.length > 0 ? "!bg-[#444]" : "!bg-transparent"}`} />
+    </div>
+  );
+}
+
+export default memo(AgentNodeComponent);
@@ -0,0 +1,302 @@
+import { Component, useMemo, type ReactNode } from "react";
+import { Brain, Bot } from "lucide-react";
+import { getToolRenderer, getToolIcon } from "./tool-renderers";
+import ChatBubble from "./tool-renderers/ChatBubble";
+import type { ToolRendererProps, AgentNode as GraphAgentNode } from "@/types/events";
+import type { TranscriptAgent, TranscriptEvent } from "@/data/serverSource";
+
+/* ---------- Error boundary so one bad event never blanks the transcript ---------- */
+class RendererErrorBoundary extends Component<
+  { toolName: string; children: ReactNode },
+  { hasError: boolean }
+> {
+  constructor(props: { toolName: string; children: ReactNode }) {
+    super(props);
+    this.state = { hasError: false };
+  }
+  static getDerivedStateFromError() {
+    return { hasError: true };
+  }
+  render() {
+    if (this.state.hasError) {
+      return (
+        <span className="text-[#555] font-semibold text-sm">
+          {this.props.toolName.replace(/_/g, " ")}
+        </span>
+      );
+    }
+    return this.props.children;
+  }
+}
+
+function SafeToolRenderer(props: ToolRendererProps) {
+  const Renderer = getToolRenderer(props.toolName);
+  return (
+    <RendererErrorBoundary toolName={props.toolName}>
+      <Renderer {...props} />
+    </RendererErrorBoundary>
+  );
+}
+
+/* ---------- Value coercion ----------
+ * args/result arrive as either a JSON object or a Python-repr string
+ * ("{'thought': '...'}"). Try JSON, then a naive python->json pass, then wrap
+ * the raw string so the fallback renderer can display it. Never throws. */
+function coerce(value: unknown): unknown {
+  if (value == null || typeof value !== "string") return value;
+  const t = value.trim();
+  if (!t) return value;
+  try {
+    return JSON.parse(t);
+  } catch {
+    /* not JSON */
+  }
+  try {
+    const jsonish = t
+      .replace(/\bNone\b/g, "null")
+      .replace(/\bTrue\b/g, "true")
+      .replace(/\bFalse\b/g, "false")
+      .replace(/'/g, '"');
+    return JSON.parse(jsonish);
+  } catch {
+    return { __raw: value };
+  }
+}
+
+function asRecord(value: unknown): Record<string, unknown> {
+  const c = coerce(value);
+  if (c && typeof c === "object" && !Array.isArray(c)) return c as Record<string, unknown>;
+  if (c == null) return {};
+  return { __raw: typeof c === "string" ? c : JSON.stringify(c) };
+}
+
+/** Numeric suffix of an event id ("tool_37" -> 37) for stable ordering. */
+function eventSeq(id: string): number {
+  const m = /(\d+)$/.exec(id);
+  return m ? parseInt(m[1], 10) : 0;
+}
+
+/** A chat event whose author is the human/user (vs an assistant "thinking"). */
+function isUserChat(event: TranscriptEvent): boolean {
+  const role = event.data?.role;
+  return event.type === "chat" && (role === "user" || role === "human");
+}
+
+/**
+ * Inter-agent message deliveries land in the recipient's session as user-role
+ * items prefixed with a header (see the engine's message formatter). They are
+ * already represented via the sending agent's tool renderer, so we never render
+ * them as chat bubbles here.
+ */
+function isInterAgentDelivery(event: TranscriptEvent): boolean {
+  return isUserChat(event) && String(event.data?.content ?? "").startsWith("[Message from ");
+}
+
+/**
+ * The reconstructed SDK session records every incoming user-role item for an
+ * agent: its initial input (the root's assembled brief, or a subagent's spawn /
+ * inherited-context prompt), inter-agent deliveries, AND genuine human steering
+ * messages sent live from the viewer or TUI. We only want the last group. Given
+ * an agent's events in order, hide the first user message (its initial input)
+ * and every inter-agent delivery; keep the rest, which are the human's live
+ * instructions, rendered as "User" bubbles.
+ */
+function hiddenUserEventIds(agentEventsInOrder: TranscriptEvent[]): Set<string> {
+  const hidden = new Set<string>();
+  let sawInitialInput = false;
+  for (const e of agentEventsInOrder) {
+    if (!isUserChat(e)) continue;
+    if (isInterAgentDelivery(e)) {
+      hidden.add(e.id);
+      continue;
+    }
+    if (!sawInitialInput) {
+      sawInitialInput = true;
+      hidden.add(e.id);
+    }
+  }
+  return hidden;
+}
+
+const STATUS_STYLE: Record<string, string> = {
+  completed: "text-emerald-400 border-emerald-500/30 bg-emerald-500/10",
+  running: "text-blue-400 border-blue-500/30 bg-blue-500/10",
+  waiting: "text-yellow-400 border-yellow-500/30 bg-yellow-500/10",
+  stopped: "text-[#aaa] border-[#333] bg-[#1a1a1a]",
+  crashed: "text-red-400 border-red-500/30 bg-red-500/10",
+  failed: "text-red-400 border-red-500/30 bg-red-500/10",
+};
+
+/** Map our engine agent statuses onto the graph node's status union. */
+function graphStatus(status: string): GraphAgentNode["status"] {
+  if (status === "completed") return "completed";
+  if (status === "running") return "running";
+  if (status === "failed" || status === "crashed") return "failed";
+  // waiting / stopped / unknown → keep the raw string; AgentNode/MiniMap fall
+  // back to a neutral gray for anything they don't explicitly style.
+  return status as GraphAgentNode["status"];
+}
+
+/**
+ * Adapt transcript agents + events into the Map<id, AgentNode> that the live
+ * AgentGraph renders: children from parent_id, tool/message counts by scanning
+ * events, and a task pulled from the spawning create_agent call where present.
+ */
+export function buildGraphAgents(
+  agents: TranscriptAgent[],
+  events: TranscriptEvent[]
+): Map<string, GraphAgentNode> {
+  const childrenOf = new Map<string, string[]>();
+  for (const a of agents) {
+    if (a.parent_id) {
+      const arr = childrenOf.get(a.parent_id) ?? [];
+      arr.push(a.id);
+      childrenOf.set(a.parent_id, arr);
+    }
+  }
+
+  const toolCount = new Map<string, number>();
+  const messageCount = new Map<string, number>();
+  // A create_agent call names the child but not its id, so map spawned tasks by
+  // agent NAME (best-effort — used only for the graph node subtitle).
+  const taskByName = new Map<string, string>();
+  for (const e of events) {
+    if (e.type === "tool") {
+      toolCount.set(e.agent_id, (toolCount.get(e.agent_id) ?? 0) + 1);
+      if (e.data?.tool_name === "create_agent") {
+        const args = asRecord(e.data.args);
+        const name = (args.name as string) ?? (args.agent_name as string) ?? "";
+        const task = (args.task as string) ?? "";
+        if (name && task) taskByName.set(name, task);
+      }
+    } else if (!isUserChat(e)) {
+      // Count only assistant messages for the graph node subtitle.
+      messageCount.set(e.agent_id, (messageCount.get(e.agent_id) ?? 0) + 1);
+    }
+  }
+
+  const map = new Map<string, GraphAgentNode>();
+  for (const a of agents) {
+    map.set(a.id, {
+      id: a.id,
+      name: a.name,
+      task: taskByName.get(a.name) ?? "",
+      status: graphStatus(a.status),
+      parentId: a.parent_id,
+      children: childrenOf.get(a.id) ?? [],
+      createdAt: a.created_at,
+      toolCount: toolCount.get(a.id) ?? 0,
+      messageCount: messageCount.get(a.id) ?? 0,
+    });
+  }
+  return map;
+}
+
+/* ---------- Per-agent transcript ---------- */
+export function AgentTranscript({
+  agent,
+  events,
+  showHeader = true,
+}: {
+  agent: TranscriptAgent;
+  events: TranscriptEvent[];
+  showHeader?: boolean;
+}) {
+  const mine = useMemo(() => {
+    const ordered = events
+      .filter((e) => e.agent_id === agent.id)
+      .sort((a, b) => eventSeq(a.id) - eventSeq(b.id));
+    const hidden = hiddenUserEventIds(ordered);
+    return ordered.filter((e) => !hidden.has(e.id));
+  }, [events, agent.id]);
+
+  const toolCount = mine.filter((e) => e.type === "tool").length;
+  const msgCount = mine.length - toolCount;
+
+  return (
+    <div>
+      {showHeader && (
+        <>
+          <div className="flex items-center gap-2 flex-wrap mb-1">
+            <span className="text-base font-semibold text-white truncate">{agent.name}</span>
+            <span
+              className={`flex-shrink-0 text-xs font-medium capitalize px-2 py-0.5 rounded-full border ${
+                STATUS_STYLE[agent.status] ?? "text-[#aaa] border-[#333] bg-[#1a1a1a]"
+              }`}
+            >
+              {agent.status}
+            </span>
+            <span className="font-mono text-xs text-[#555]">{agent.id}</span>
+          </div>
+          <p className="text-xs text-[#666] mb-4">
+            {msgCount} message{msgCount === 1 ? "" : "s"} · {toolCount} tool call
+            {toolCount === 1 ? "" : "s"}
+          </p>
+        </>
+      )}
+
+      {mine.length === 0 ? (
+        <p className="text-sm text-[#666]">No recorded activity for this agent.</p>
+      ) : (
+        <div className="py-1">
+          {mine.map((event, i) => {
+            const isLast = i === mine.length - 1;
+            const isTool = event.type === "tool";
+            const toolName = isTool ? String(event.data?.tool_name ?? "tool") : "";
+            const role = !isTool ? String(event.data?.role ?? "assistant") : "";
+
+            let Icon;
+            let iconColor: string;
+            if (isTool) {
+              const meta = getToolIcon(toolName);
+              Icon = meta.icon;
+              iconColor = meta.color;
+            } else {
+              const isUser = role === "user" || role === "human";
+              Icon = isUser ? Bot : Brain;
+              iconColor = isUser ? "text-blue-400" : "text-purple-400";
+            }
+
+            const status = isTool ? String(event.data?.status ?? "completed") : "completed";
+
+            return (
+              <div key={event.id} className="flex gap-3">
+                <div className="flex flex-col items-center shrink-0">
+                  <div
+                    className={`w-[30px] h-[30px] rounded-full bg-black border flex items-center justify-center shrink-0 ${
+                      isTool && status === "running"
+                        ? "border-blue-500/40 animate-pulse"
+                        : isTool && status === "failed"
+                          ? "border-red-500/30"
+                          : "border-[#222]"
+                    }`}
+                  >
+                    <Icon className={`w-3.5 h-3.5 ${iconColor}`} />
+                  </div>
+                  {!isLast && <div className="w-px flex-1 bg-[#1a1a1a] mt-1" />}
+                </div>
+                <div className="flex-1 min-w-0 pt-[5px] pb-6">
+                  {isTool ? (
+                    <SafeToolRenderer
+                      toolName={toolName}
+                      args={asRecord(event.data?.args)}
+                      result={coerce(event.data?.result) ?? null}
+                      status={
+                        status as ToolRendererProps["status"]
+                      }
+                    />
+                  ) : (
+                    <ChatBubble
+                      role={role}
+                      content={String(event.data?.content ?? "")}
+                    />
+                  )}
+                </div>
+              </div>
+            );
+          })}
+        </div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,70 @@
+"use client";
+
+function SkeletonNode({ w = 24 }: { w?: number }) {
+  return (
+    <div className="w-[180px] h-[72px] rounded-lg border border-[#222] bg-[#0a0a0a] px-3 py-2 shrink-0">
+      <div className="flex items-center gap-2 mb-1.5">
+        <div className="w-2 h-2 rounded-full bg-[#2a2a2a]" />
+        <div className="h-3 rounded bg-[#252525]" style={{ width: `${w * 4}px` }} />
+      </div>
+      <div className="h-2 w-28 rounded bg-[#1e1e1e] mb-1.5" />
+      <div className="flex gap-3">
+        <div className="h-2 w-8 rounded bg-[#1e1e1e]" />
+        <div className="h-2 w-8 rounded bg-[#1e1e1e]" />
+      </div>
+    </div>
+  );
+}
+
+function VLine() {
+  return <div className="w-px h-6 bg-[#2a2a2a]" />;
+}
+
+function HBranch({ count }: { count: number }) {
+  return (
+    <div className="relative flex justify-center">
+      <div className="absolute top-0 h-px bg-[#2a2a2a]" style={{ width: `${(count - 1) * 220}px` }} />
+    </div>
+  );
+}
+
+export default function GraphSkeleton() {
+  return (
+    <div className="h-full bg-black overflow-hidden">
+      <div className="flex flex-col items-center pt-10 animate-pulse">
+        <SkeletonNode w={20} />
+        <VLine />
+        <HBranch count={3} />
+        <div className="flex gap-10">
+          {[18, 22, 16].map((w, i) => (
+            <div key={i} className="flex flex-col items-center">
+              <VLine />
+              <SkeletonNode w={w} />
+            </div>
+          ))}
+        </div>
+        <div className="flex gap-10 w-full justify-center">
+          <div className="flex flex-col items-center">
+            <VLine />
+            <HBranch count={2} />
+            <div className="flex gap-10">
+              {[14, 20].map((w, i) => (
+                <div key={i} className="flex flex-col items-center">
+                  <VLine />
+                  <SkeletonNode w={w} />
+                </div>
+              ))}
+            </div>
+          </div>
+          <div className="flex flex-col items-center">
+            <VLine />
+            <SkeletonNode w={18} />
+            <VLine />
+            <SkeletonNode w={12} />
+          </div>
+          <div className="w-[180px]" />
+        </div>
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,315 @@
+import {
+  useCallback,
+  useEffect,
+  useLayoutEffect,
+  useMemo,
+  useRef,
+  useState,
+} from "react";
+import { ArrowUp, ChevronDown, ChevronUp, Loader2, Sparkles } from "lucide-react";
+import { steerAgent, type TranscriptAgent } from "@/data/serverSource";
+import { track } from "@/lib/cta";
+import { cn } from "@/lib/utils";
+
+const ROOT_TARGET_VALUE = "__root__";
+
+interface ScanPromptComposerProps {
+  /** All agents in the run; used to resolve the root and running children. */
+  agents: TranscriptAgent[];
+  /**
+   * Single-agent (modal) mode: pins the composer to one agent and shows a
+   * static "Target: <name>" pill instead of the dropdown. Omit for the
+   * multi-agent graph variant.
+   */
+  fixedAgentId?: string;
+  className?: string;
+}
+
+/**
+ * Faithful port of the pro app's ScanPromptComposer for the local viewer.
+ * Collapsed by default into a "Guide the agent" pill; expands into a card with
+ * an auto-resizing textarea and a target control. The viewer's steering is
+ * immediate (no Enterprise lock, no bridge-connecting state), so this is only
+ * rendered by callers when steering is available. Sends via steerAgent, which
+ * requires a concrete agent id, so "Root agent" resolves to the root agent's id.
+ */
+export function ScanPromptComposer({
+  agents,
+  fixedAgentId,
+  className,
+}: ScanPromptComposerProps) {
+  const textareaRef = useRef<HTMLTextAreaElement>(null);
+  const [expanded, setExpanded] = useState(false);
+  const [focused, setFocused] = useState(false);
+  const [value, setValue] = useState("");
+  const [sending, setSending] = useState(false);
+  const [feedback, setFeedback] = useState<string | null>(null);
+
+  const isModal = fixedAgentId != null;
+
+  // Root = the agent with no parent; fall back to the first agent.
+  const rootAgent = useMemo(
+    () => agents.find((a) => !a.parent_id) ?? agents[0] ?? null,
+    [agents]
+  );
+
+  // Multi-agent dropdown options: running child agents plus Root (added in JSX).
+  const targetOptions = useMemo(
+    () => agents.filter((a) => a.parent_id && a.status === "running"),
+    [agents]
+  );
+
+  // Selected target for the multi-agent variant. ROOT sentinel by default.
+  const [selectedTarget, setSelectedTarget] = useState<string>(ROOT_TARGET_VALUE);
+  const [menuOpen, setMenuOpen] = useState(false);
+
+  // If the selected child target disappears (finished), fall back to Root.
+  useEffect(() => {
+    if (
+      selectedTarget !== ROOT_TARGET_VALUE &&
+      !targetOptions.some((a) => a.id === selectedTarget)
+    ) {
+      setSelectedTarget(ROOT_TARGET_VALUE);
+    }
+  }, [selectedTarget, targetOptions]);
+
+  // Resolve the concrete agent id + display name for the current target.
+  const { targetId, targetName } = useMemo(() => {
+    if (isModal) {
+      const agent = agents.find((a) => a.id === fixedAgentId) ?? null;
+      return {
+        targetId: fixedAgentId ?? null,
+        targetName: agent?.name ?? "this agent",
+      };
+    }
+    if (selectedTarget === ROOT_TARGET_VALUE) {
+      return {
+        targetId: rootAgent?.id ?? null,
+        targetName: "Root agent",
+      };
+    }
+    const agent = agents.find((a) => a.id === selectedTarget) ?? null;
+    return {
+      targetId: agent?.id ?? rootAgent?.id ?? null,
+      targetName: agent?.name ?? "Root agent",
+    };
+  }, [agents, fixedAgentId, isModal, rootAgent, selectedTarget]);
+
+  const empty = value.trim().length === 0;
+
+  // Grow the textarea with its content, capped by max-h via CSS.
+  useLayoutEffect(() => {
+    const el = textareaRef.current;
+    if (!el) return;
+    el.style.height = "auto";
+    el.style.height = `${el.scrollHeight}px`;
+  }, [value]);
+
+  const handleExpand = useCallback(() => {
+    setExpanded(true);
+    requestAnimationFrame(() => textareaRef.current?.focus());
+  }, []);
+
+  const handleCollapse = useCallback(() => {
+    setExpanded(false);
+    setFocused(false);
+    setMenuOpen(false);
+  }, []);
+
+  const handleSend = useCallback(async () => {
+    if (sending) return;
+    const message = value.trim();
+    if (!message || !targetId) return;
+
+    setSending(true);
+    setFeedback(null);
+    const name = targetName;
+    const res = await steerAgent(targetId, message);
+    setSending(false);
+    if (res.ok) {
+      setValue("");
+      setFeedback(`Sent to ${name}`);
+      track("agent_steered");
+    } else if (res.error === "not_delivered") {
+      setFeedback("Could not reach that agent (it may have finished).");
+    } else {
+      setFeedback("Could not send that message. Try again.");
+    }
+  }, [sending, value, targetId, targetName]);
+
+  if (!expanded) {
+    return (
+      <button
+        type="button"
+        onClick={handleExpand}
+        className={cn(
+          "mt-4 flex w-full items-center justify-between gap-3 rounded-2xl border border-white/[0.08] bg-[#050505] px-5 py-3 text-left transition-colors duration-300 hover:border-white/[0.12] focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-white/20",
+          className
+        )}
+        aria-expanded={false}
+        aria-label="Expand live prompt composer"
+      >
+        <div className="flex min-w-0 items-center gap-2">
+          <Sparkles className="h-4 w-4 shrink-0 text-[#666]" />
+          <span className="truncate text-sm font-medium text-white">Guide the agent</span>
+        </div>
+        <ChevronUp className="h-4 w-4 shrink-0 text-[#777]" />
+      </button>
+    );
+  }
+
+  return (
+    <div
+      className={cn(
+        "mt-4 rounded-2xl border border-white/[0.08] bg-[#050505] overflow-hidden transition-colors duration-300",
+        focused ? "border-white/[0.18]" : "hover:border-white/[0.12]",
+        className
+      )}
+    >
+      <div className="flex items-center justify-between gap-3 border-b border-white/[0.06] px-5 py-3">
+        <div className="min-w-0">
+          <div className="flex items-center gap-2">
+            <Sparkles className="h-4 w-4 text-[#666]" />
+            <p className="text-sm font-medium text-white">Live prompt</p>
+          </div>
+          <p className="mt-0.5 text-xs text-[#777]">Connected</p>
+        </div>
+        <div className="flex shrink-0 items-center gap-2">
+          {isModal ? (
+            <div className="rounded-full border border-white/[0.08] bg-white/[0.03] px-3 py-1 text-xs text-[#aaa]">
+              Target: <span className="text-white">{targetName}</span>
+            </div>
+          ) : (
+            <div className="flex items-center gap-1.5">
+              <span className="text-xs text-[#aaa]">Target:</span>
+              <div className="relative">
+                <button
+                  type="button"
+                  onClick={() => setMenuOpen((o) => !o)}
+                  onBlur={() => requestAnimationFrame(() => setMenuOpen(false))}
+                  className="inline-flex h-7 items-center gap-1 rounded-full border border-white/[0.08] bg-white/[0.03] px-3 text-xs text-white transition-colors hover:border-white/[0.16] focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-white/20"
+                  aria-haspopup="listbox"
+                  aria-expanded={menuOpen}
+                >
+                  <span className="max-w-[140px] truncate">{targetName}</span>
+                  <ChevronDown className="h-3.5 w-3.5 text-[#999]" />
+                </button>
+                {menuOpen && (
+                  <div
+                    className="absolute right-0 z-10 mt-1 min-w-[160px] overflow-hidden rounded-lg border border-[#333] bg-[#0a0a0a] py-1 shadow-xl"
+                    role="listbox"
+                  >
+                    <TargetMenuItem
+                      label="Root agent"
+                      active={selectedTarget === ROOT_TARGET_VALUE}
+                      onSelect={() => {
+                        setSelectedTarget(ROOT_TARGET_VALUE);
+                        setMenuOpen(false);
+                      }}
+                    />
+                    {targetOptions.map((option) => (
+                      <TargetMenuItem
+                        key={option.id}
+                        label={option.name}
+                        active={selectedTarget === option.id}
+                        onSelect={() => {
+                          setSelectedTarget(option.id);
+                          setMenuOpen(false);
+                        }}
+                      />
+                    ))}
+                  </div>
+                )}
+              </div>
+            </div>
+          )}
+          <button
+            type="button"
+            onClick={handleCollapse}
+            className="inline-flex h-7 w-7 items-center justify-center rounded-full text-[#777] transition-colors hover:bg-white/[0.06] hover:text-white focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-white/20"
+            aria-label="Collapse live prompt composer"
+          >
+            <ChevronDown className="h-4 w-4" />
+          </button>
+        </div>
+      </div>
+
+      <div className="px-5 pt-4 pb-3">
+        <textarea
+          ref={textareaRef}
+          rows={1}
+          value={value}
+          onChange={(e) => setValue(e.target.value)}
+          onFocus={() => setFocused(true)}
+          onBlur={() => setFocused(false)}
+          onKeyDown={(e) => {
+            if (e.key === "Enter" && !e.shiftKey) {
+              e.preventDefault();
+              void handleSend();
+            }
+          }}
+          placeholder="Send a live prompt to the running scan…"
+          maxLength={4000}
+          disabled={sending}
+          className="block w-full resize-none border-0 bg-transparent p-0 text-[15px] leading-6 text-white placeholder:text-[#444] focus:outline-none disabled:opacity-60 max-h-[160px] overflow-y-auto"
+        />
+      </div>
+
+      <div className="flex items-center justify-between gap-3 px-4 pb-4">
+        <div className="text-xs text-[#666]">{feedback ?? "Press Enter to send."}</div>
+        <button
+          type="button"
+          onClick={(event) => {
+            event.stopPropagation();
+            void handleSend();
+          }}
+          disabled={sending || empty}
+          className={cn(
+            "inline-flex h-10 min-w-[112px] items-center justify-center gap-2 rounded-full px-4 text-sm font-medium transition-colors",
+            sending || empty
+              ? "bg-white/[0.08] text-[#666]"
+              : "bg-white text-black hover:bg-neutral-200"
+          )}
+        >
+          {sending ? (
+            <Loader2 className="h-4 w-4 animate-spin" />
+          ) : (
+            <ArrowUp className="h-4 w-4" strokeWidth={2.5} />
+          )}
+          <span>Send prompt</span>
+        </button>
+      </div>
+    </div>
+  );
+}
+
+function TargetMenuItem({
+  label,
+  active,
+  onSelect,
+}: {
+  label: string;
+  active: boolean;
+  onSelect: () => void;
+}) {
+  return (
+    <button
+      type="button"
+      role="option"
+      aria-selected={active}
+      // onMouseDown so the click lands before the trigger's onBlur closes the menu.
+      onMouseDown={(e) => {
+        e.preventDefault();
+        onSelect();
+      }}
+      className={cn(
+        "block w-full truncate px-3 py-1.5 text-left text-xs transition-colors hover:bg-white/[0.06]",
+        active ? "text-white" : "text-[#aaa]"
+      )}
+    >
+      {label}
+    </button>
+  );
+}
+
+export default ScanPromptComposer;
@@ -0,0 +1,92 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { TruncatedText } from "./ToolCard";
+
+export default function AgentCommsRenderer({ toolName, args }: ToolRendererProps) {
+  if (toolName === "create_agent") {
+    const name = (args.name as string) ?? (args.agent_name as string) ?? "";
+    const task = (args.task as string) ?? "";
+    return (
+      <div>
+        <div className="flex items-center gap-2">
+          <span className="text-cyan-400/80 font-semibold text-sm">spawning</span>
+          {name && <span className="text-cyan-400 font-semibold text-sm">{name}</span>}
+        </div>
+        {task && <div className="mt-1.5"><TruncatedText text={task} maxLines={15} /></div>}
+      </div>
+    );
+  }
+
+  if (toolName === "agent_finish") {
+    const summary = (args.result_summary as string) ?? "";
+    const success = args.success as boolean | undefined;
+    const rawFindings = args.findings;
+    const findings = Array.isArray(rawFindings) ? rawFindings as string[] : undefined;
+    return (
+      <div>
+        <span className={`font-semibold text-sm ${success === false ? "text-red-400/80" : "text-emerald-400/80"}`}>
+          {success === false ? "Agent failed" : "Agent completed"}
+        </span>
+        {summary && <div className="mt-1.5"><TruncatedText text={summary} maxLines={20} /></div>}
+        {findings && findings.length > 0 && (
+          <div className="mt-1.5 space-y-0.5">
+            {findings.map((f, i) => (
+              <div key={i} className="text-[13px] text-[#888]"><span className="text-red-400/50 mr-1">•</span>{typeof f === "string" ? f : JSON.stringify(f)}</div>
+            ))}
+          </div>
+        )}
+      </div>
+    );
+  }
+
+  if (toolName === "send_message_to_agent") {
+    const message = (args.message as string) ?? "";
+    const agentId = (args.target_agent_id as string) ?? (args.agent_id as string) ?? "";
+    return (
+      <div>
+        <div className="flex items-center gap-2">
+          <span className="text-cyan-400/80 font-semibold text-sm">message</span>
+          {agentId && <span className="text-[#888] text-[13px]">to {agentId.slice(0, 16)}</span>}
+        </div>
+        {message && <div className="mt-1.5"><TruncatedText text={message} maxLines={20} /></div>}
+      </div>
+    );
+  }
+
+  if (toolName === "wait_for_message") {
+    const reason = (args.reason as string) ?? "";
+    return (
+      <div className="flex items-center gap-2">
+        <span className="text-cyan-400/80 font-semibold text-sm">waiting</span>
+        {reason && <span className="text-[#888] text-[13px] truncate">{reason}</span>}
+      </div>
+    );
+  }
+
+  if (toolName === "stop_agent") {
+    const targetAgentId = (args.target_agent_id as string) ?? "";
+    const cascade = args.cascade !== false;
+    const reason = (args.reason as string) ?? "";
+    return (
+      <div>
+        <div className="flex items-center gap-2 flex-wrap">
+          <span className="text-red-400/80 font-semibold text-sm">stopping</span>
+          {targetAgentId && <span className="text-[#888] text-[13px]">{targetAgentId.slice(0, 16)}</span>}
+          {cascade && <span className="text-[#555] text-[13px] italic">+ descendants</span>}
+        </div>
+        {reason && <div className="mt-1.5 text-[#888] text-[13px]">{reason}</div>}
+      </div>
+    );
+  }
+
+  if (toolName === "view_agent_graph") {
+    return (
+      <span className="text-cyan-400/80 font-semibold text-sm">viewing agents graph</span>
+    );
+  }
+
+  return (
+    <span className="text-cyan-400/80 font-semibold text-sm">{toolName.replace(/_/g, " ")}</span>
+  );
+}
@@ -0,0 +1,124 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { shortPath } from "./utils";
+
+const DIFF_PREVIEW_LINES = 30;
+
+const BEGIN_PATCH = "*** Begin Patch";
+const END_PATCH = "*** End Patch";
+const ADD_FILE = "*** Add File: ";
+const UPDATE_FILE = "*** Update File: ";
+const DELETE_FILE = "*** Delete File: ";
+
+const OP_LABEL: Record<string, string> = { add: "create", update: "edit", delete: "delete" };
+
+interface PatchOp {
+  kind: "add" | "update" | "delete";
+  path: string;
+  oldLines: string[];
+  newLines: string[];
+}
+
+/** apply_patch args arrive as {patch: text} (chat-completions FunctionTool) or
+ *  {input: text} (CustomTool). Mirrors the OSS `_extract_patch_text`. */
+function extractPatchText(args: Record<string, unknown>): string {
+  const raw = args.patch;
+  if (typeof raw === "string") return raw;
+  if (raw && typeof raw === "object" && typeof (raw as Record<string, unknown>).patch === "string") {
+    return (raw as Record<string, string>).patch;
+  }
+  return typeof args.input === "string" ? args.input : "";
+}
+
+/** Parse V4A patch text into per-file operations (mirrors `_parse_patch_operations`). */
+function parsePatchOperations(patchText: string): PatchOp[] {
+  const ops: PatchOp[] = [];
+  let current: PatchOp | null = null;
+
+  const flush = () => {
+    if (current) ops.push(current);
+    current = null;
+  };
+
+  for (const line of patchText.split("\n")) {
+    if (line === BEGIN_PATCH || line === END_PATCH) continue;
+    if (line.startsWith(ADD_FILE)) {
+      flush();
+      current = { kind: "add", path: line.slice(ADD_FILE.length).trim(), oldLines: [], newLines: [] };
+    } else if (line.startsWith(UPDATE_FILE)) {
+      flush();
+      current = { kind: "update", path: line.slice(UPDATE_FILE.length).trim(), oldLines: [], newLines: [] };
+    } else if (line.startsWith(DELETE_FILE)) {
+      flush();
+      current = { kind: "delete", path: line.slice(DELETE_FILE.length).trim(), oldLines: [], newLines: [] };
+    } else if (current?.kind === "update") {
+      if (line.startsWith("@@")) continue;
+      if (line.startsWith("-") && !line.startsWith("---")) current.oldLines.push(line.slice(1));
+      else if (line.startsWith("+") && !line.startsWith("+++")) current.newLines.push(line.slice(1));
+    } else if (current?.kind === "add") {
+      if (line.startsWith("+")) current.newLines.push(line.slice(1));
+      else if (line.trim()) current.newLines.push(line);
+    }
+  }
+  flush();
+  return ops;
+}
+
+function Operation({ op }: { op: PatchOp }) {
+  const label = OP_LABEL[op.kind] ?? "file";
+  const total = op.oldLines.length + op.newLines.length;
+  const truncated = total > DIFF_PREVIEW_LINES;
+  const oldBudget = truncated && total > 0 ? Math.round(DIFF_PREVIEW_LINES * (op.oldLines.length / total)) : op.oldLines.length;
+  const newBudget = truncated ? DIFF_PREVIEW_LINES - oldBudget : op.newLines.length;
+
+  return (
+    <div>
+      <div className="flex items-baseline gap-2">
+        <span className="text-sky-400/80 font-semibold text-sm shrink-0">{label}</span>
+        {op.path && <span className="text-[#888] font-mono text-[13px] break-all">{shortPath(op.path)}</span>}
+      </div>
+      {(op.oldLines.length > 0 || op.newLines.length > 0) && (
+        <div className="font-mono text-[13px] leading-relaxed mt-1.5">
+          {op.oldLines.slice(0, oldBudget).map((line, i) => (
+            <div key={`o${i}`} className="text-red-400/60">
+              <span className="select-none text-red-400/30 mr-1">-</span>{line}
+            </div>
+          ))}
+          {op.newLines.slice(0, newBudget).map((line, i) => (
+            <div key={`n${i}`} className="text-emerald-400/60">
+              <span className="select-none text-emerald-400/30 mr-1">+</span>{line}
+            </div>
+          ))}
+          {truncated && <div className="text-[#444] mt-0.5">... {total - DIFF_PREVIEW_LINES} more lines</div>}
+        </div>
+      )}
+    </div>
+  );
+}
+
+export default function ApplyPatchRenderer({ args, result, status }: ToolRendererProps) {
+  const ops = parsePatchOperations(extractPatchText(args));
+
+  if (ops.length === 0) {
+    return (
+      <div>
+        <span className="text-sky-400/80 font-semibold text-sm">patch</span>
+        {status === "failed" && typeof result === "string" && result.trim() && (
+          <div className="text-red-400/70 text-[13px] mt-1">{result.trim()}</div>
+        )}
+      </div>
+    );
+  }
+
+  return (
+    <div className="space-y-2">
+      {ops.map((op, i) => (
+        <Operation key={i} op={op} />
+      ))}
+      {status === "failed" && typeof result === "string" && result.trim() && (
+        <div className="text-red-400/70 text-[13px]">{result.trim()}</div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,110 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { SyntaxBlock } from "./ToolCard";
+
+const SIMPLE_ACTIONS: Record<string, string> = {
+  back: "going back in browser history",
+  forward: "going forward in browser history",
+  scroll_down: "scrolling down",
+  scroll_up: "scrolling up",
+  refresh: "refreshing",
+  close_tab: "closing tab",
+  switch_tab: "switching tab",
+  list_tabs: "listing tabs",
+  view_source: "viewing page source",
+  get_console_logs: "getting console logs",
+  screenshot: "taking screenshot",
+  wait: "waiting...",
+  close: "closing",
+};
+
+const CLICK_ACTIONS: Record<string, string> = {
+  click: "clicking",
+  double_click: "double clicking",
+  hover: "hovering",
+};
+
+function UrlLabel({ prefix, url, suffix }: { prefix: string; url?: string; suffix?: string }) {
+  return (
+    <span className="text-[#888] text-[13px]">
+      {prefix}
+      {url && (
+        <a
+          href={url}
+          target="_blank"
+          rel="noopener noreferrer"
+          className="text-cyan-400/80 hover:underline"
+        >
+          {url}
+        </a>
+      )}
+      {suffix}
+    </span>
+  );
+}
+
+function describeAction(args: Record<string, unknown>): React.ReactNode {
+  const action = (args.action as string) ?? "";
+  const url = (args.url as string) ?? undefined;
+
+  // Simple actions (no extra args)
+  if (action in SIMPLE_ACTIONS) return SIMPLE_ACTIONS[action];
+
+  // URL actions: launch, goto, new_tab
+  if (action === "launch") {
+    if (!url) return "launching";
+    return <UrlLabel prefix="launching " url={url} />;
+  }
+  if (action === "goto" || action === "navigate") {
+    return <UrlLabel prefix="navigating to " url={url} />;
+  }
+  if (action === "new_tab") {
+    return <UrlLabel prefix="opening tab " url={url} />;
+  }
+
+  // Click actions
+  if (action in CLICK_ACTIONS) return CLICK_ACTIONS[action];
+
+  // Type
+  if (action === "type") {
+    const text = ((args.text as string) ?? "").slice(0, 40);
+    return `typing "${text}"`;
+  }
+
+  // Key press
+  if (action === "press_key" || action === "key_press") {
+    return `pressing key ${(args.key as string) ?? ""}`;
+  }
+
+  // Save PDF
+  if (action === "save_pdf" || action === "save_as_pdf") {
+    const path = (args.file_path as string) ?? "";
+    return `saving PDF${path ? ` to ${path}` : ""}`;
+  }
+
+  // Execute JS — description only, code shown separately
+  if (action === "execute_js") return "executing javascript";
+
+  return action || "browser action";
+}
+
+export default function BrowserRenderer({ args }: ToolRendererProps) {
+  const action = (args.action as string) ?? "";
+  const jsCode = action === "execute_js"
+    ? ((args.js_code as string) ?? (args.code as string) ?? "")
+    : "";
+  const description = describeAction(args);
+
+  return (
+    <div>
+      <div className="flex items-center gap-2 min-w-0">
+        <span className="text-blue-400/80 font-semibold text-sm shrink-0">Browser</span>
+        <span className="min-w-0 truncate text-[#888] text-[13px]">
+          {typeof description === "string" ? description : description}
+        </span>
+      </div>
+      {jsCode && <SyntaxBlock code={jsCode} language="javascript" collapsible />}
+    </div>
+  );
+}
@@ -0,0 +1,25 @@
+"use client";
+
+import { TruncatedText } from "./ToolCard";
+
+interface ChatBubbleProps {
+  role: string;
+  content: string;
+}
+
+const MAX_LINES = 30;
+
+export default function ChatBubble({ role, content }: ChatBubbleProps) {
+  const isUser = role === "user" || role === "human";
+
+  return (
+    <div>
+      <span className={`font-semibold text-sm ${isUser ? "text-blue-400/80" : "text-purple-400/80"}`}>
+        {isUser ? "User" : "Thinking"}
+      </span>
+      <div className="mt-1.5 italic text-[#888]">
+        <TruncatedText text={content} maxLines={MAX_LINES} />
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,36 @@
+import type { ToolRendererProps } from "@/types/events";
+import { CodeBlock } from "./ToolCard";
+
+/**
+ * Generic renderer for tool names without a dedicated family renderer. Shows the
+ * humanized tool name plus a pretty-printed dump of args/result. Tolerates the
+ * server sending args/result as either a parsed object or an unparseable
+ * Python-repr string (which arrives here wrapped as { __raw }); never crashes.
+ */
+function pretty(value: unknown): string | null {
+  if (value == null) return null;
+  if (typeof value === "string") return value.trim() ? value : null;
+  if (typeof value === "object") {
+    const rec = value as Record<string, unknown>;
+    if (typeof rec.__raw === "string") return rec.__raw;
+    if (Object.keys(rec).length === 0) return null;
+    try {
+      return JSON.stringify(value, null, 2);
+    } catch {
+      return String(value);
+    }
+  }
+  return String(value);
+}
+
+export default function FallbackRenderer({ toolName, args, result }: ToolRendererProps) {
+  const argsText = pretty(args);
+  const resultText = pretty(result);
+  return (
+    <div>
+      <span className="text-[#888] font-semibold text-sm">{toolName.replace(/_/g, " ")}</span>
+      {argsText && <CodeBlock className="text-[#777]">{argsText}</CodeBlock>}
+      {resultText && <CodeBlock className="text-[#666]">{resultText}</CodeBlock>}
+    </div>
+  );
+}
@@ -0,0 +1,65 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { shortPath } from "./utils";
+
+const DIFF_PREVIEW_LINES = 30;
+
+export default function FileEditRenderer({ toolName, args }: ToolRendererProps) {
+  const filePath = (args.path as string) ?? (args.file_path as string) ?? "";
+  const command = (args.command as string) ?? "";
+  const oldStr = (args.old_str as string) ?? "";
+  const newStr = (args.new_str as string) ?? "";
+  const regex = (args.regex as string) ?? "";
+
+  let label: string;
+  if (toolName === "list_files") label = "list";
+  else if (toolName === "search_files") label = "search";
+  else if (command === "view") label = "view";
+  else if (command === "create") label = "create";
+  else if (command === "str_replace") label = "edit";
+  else if (command === "undo_edit") label = "undo";
+  else if (command === "insert") label = "insert";
+  else label = "file";
+
+  const pathDisplay = filePath ? shortPath(filePath) : "";
+  const regexDisplay = regex ? ` /${regex}/` : "";
+
+  const oldLines = oldStr ? oldStr.split("\n") : [];
+  const newLines = newStr ? newStr.split("\n") : [];
+  const totalLines = oldLines.length + newLines.length;
+  const truncated = totalLines > DIFF_PREVIEW_LINES;
+
+  // If truncated, split the budget proportionally
+  const oldBudget = truncated ? Math.round(DIFF_PREVIEW_LINES * (oldLines.length / totalLines)) : oldLines.length;
+  const newBudget = truncated ? DIFF_PREVIEW_LINES - oldBudget : newLines.length;
+
+  return (
+    <div>
+      <div className="flex items-baseline gap-2">
+        <span className="text-sky-400/80 font-semibold text-sm shrink-0">{label}</span>
+        {pathDisplay && <span className="text-[#888] font-mono text-[13px] break-all">{pathDisplay}</span>}
+      </div>
+      {regexDisplay && (
+        <div className="text-purple-400/60 font-mono text-[13px] break-all mt-0.5">{regexDisplay}</div>
+      )}
+      {(oldStr || newStr) && (
+        <div className="font-mono text-[13px] leading-relaxed mt-1.5">
+          {oldLines.slice(0, oldBudget).map((line, i) => (
+            <div key={`o${i}`} className="text-red-400/60">
+              <span className="select-none text-red-400/30 mr-1">-</span>{line}
+            </div>
+          ))}
+          {newLines.slice(0, newBudget).map((line, i) => (
+            <div key={`n${i}`} className="text-emerald-400/60">
+              <span className="select-none text-emerald-400/30 mr-1">+</span>{line}
+            </div>
+          ))}
+          {truncated && (
+            <div className="text-[#444] mt-0.5">... {totalLines - DIFF_PREVIEW_LINES} more lines</div>
+          )}
+        </div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,32 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { TruncatedText } from "./ToolCard";
+
+export default function FinishRenderer({ args }: ToolRendererProps) {
+  const executiveSummary = (args.executive_summary as string) ?? "";
+  const methodology = (args.methodology as string) ?? "";
+  const technicalAnalysis = (args.technical_analysis as string) ?? "";
+  const recommendations = (args.recommendations as string) ?? "";
+
+  return (
+    <div className="space-y-3">
+      <span className="text-emerald-400/80 font-semibold text-sm">Penetration test completed</span>
+      {executiveSummary && (
+        <div><span className="text-emerald-400/60 text-sm font-semibold">Executive Summary</span><div className="mt-1"><TruncatedText text={executiveSummary} maxLines={25} /></div></div>
+      )}
+      {methodology && (
+        <div><span className="text-emerald-400/60 text-sm font-semibold">Methodology</span><div className="mt-1"><TruncatedText text={methodology} maxLines={25} /></div></div>
+      )}
+      {technicalAnalysis && (
+        <div><span className="text-emerald-400/60 text-sm font-semibold">Technical Analysis</span><div className="mt-1"><TruncatedText text={technicalAnalysis} maxLines={25} /></div></div>
+      )}
+      {recommendations && (
+        <div><span className="text-emerald-400/60 text-sm font-semibold">Recommendations</span><div className="mt-1"><TruncatedText text={recommendations} maxLines={25} /></div></div>
+      )}
+      {!executiveSummary && !methodology && !technicalAnalysis && !recommendations && (
+        <div className="text-[#555] text-xs">Generating final report...</div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,20 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+
+export default function LoadSkillRenderer({ args }: ToolRendererProps) {
+  const requestedRaw = (args.skills as string) ?? "";
+  const requestedSkills = requestedRaw
+    .split(",")
+    .map((skill) => skill.trim())
+    .filter(Boolean);
+
+  return (
+    <div className="flex items-center gap-2">
+      <span className="text-emerald-400/80 font-semibold text-sm">Loading skill</span>
+      {requestedSkills.length > 0 && (
+        <span className="text-[#888] text-[13px]">{requestedSkills.join(", ")}</span>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,24 @@
+"use client";
+
+import ReactMarkdown from "react-markdown";
+import remarkGfm from "remark-gfm";
+import { rehypeCodeMeta, mdComponents } from "@/components/vulnerability/MdCodeBlock";
+
+interface MarkdownProps {
+  text: string;
+  className?: string;
+}
+
+export default function Markdown({ text, className = "" }: MarkdownProps) {
+  return (
+    <div className={`prose-markdown ${className}`}>
+      <ReactMarkdown
+        remarkPlugins={[remarkGfm]}
+        rehypePlugins={[rehypeCodeMeta]}
+        components={mdComponents}
+      >
+        {text}
+      </ReactMarkdown>
+    </div>
+  );
+}
@@ -0,0 +1,87 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import Markdown from "./Markdown";
+
+export default function NotesRenderer({ toolName, args, result }: ToolRendererProps) {
+  if (toolName === "create_note") {
+    const title = (args.title as string) ?? "";
+    const content = (args.content as string) ?? "";
+    const category = (args.category as string) ?? "general";
+    return (
+      <div>
+        <div className="flex items-center gap-2">
+          <span className="text-amber-400/80 font-semibold text-sm">note</span>
+          <span className="text-[#555] text-[13px]">({category})</span>
+        </div>
+        {title && <div className="mt-1.5 text-[#999] text-[13px]">{title}</div>}
+        {content && <div className="mt-1"><Markdown text={content} /></div>}
+      </div>
+    );
+  }
+
+  if (toolName === "delete_note") {
+    return <span className="text-amber-400/80 font-semibold text-sm">note removed</span>;
+  }
+
+  if (toolName === "update_note") {
+    const title = (args.title as string) ?? "";
+    const content = (args.content as string) ?? "";
+    return (
+      <div>
+        <span className="text-amber-400/80 font-semibold text-sm">note updated</span>
+        {title && <div className="mt-1.5 text-[#999] text-[13px]">{title}</div>}
+        {content && <div className="mt-1"><Markdown text={content} /></div>}
+      </div>
+    );
+  }
+
+  if (toolName === "get_note") {
+    const res = result as Record<string, unknown> | null;
+    const note = res && typeof res === "object" && res.success
+      ? (res.note as Record<string, string> | undefined)
+      : undefined;
+    return (
+      <div>
+        <span className="text-amber-400/80 font-semibold text-sm">note read</span>
+        {note && (
+          <>
+            <div className="mt-1.5 text-[#999] text-[13px]">
+              {note.title ?? "(untitled)"}
+              <span className="text-[#555] ml-1">({note.category ?? "general"})</span>
+            </div>
+            {note.content && <div className="mt-1"><Markdown text={note.content} /></div>}
+          </>
+        )}
+      </div>
+    );
+  }
+
+  if (toolName === "list_notes") {
+    const res = result as Record<string, unknown> | null;
+    let notes: Array<Record<string, string>> = [];
+    if (res && typeof res === "object" && res.success) {
+      const rawNotes = res.notes;
+      notes = Array.isArray(rawNotes) ? rawNotes as Array<Record<string, string>> : [];
+    }
+    return (
+      <div>
+        <span className="text-amber-400/80 font-semibold text-sm">notes</span>
+        {notes.length > 0 ? (
+          <div className="mt-1.5 space-y-0.5">
+            {notes.map((n, i) => (
+              <div key={i} className="text-[13px]">
+                <span className="text-[#555] mr-1">-</span>
+                <span className="text-[#999]">{n.title ?? "(untitled)"}</span>
+                <span className="text-[#555] ml-1">({n.category ?? "general"})</span>
+                {n.content && <div className="ml-3"><Markdown text={n.content} /></div>}
+              </div>
+            ))}
+          </div>
+        ) : <div className="mt-1 text-[#555] text-xs">No notes</div>}
+      </div>
+    );
+  }
+
+  return <span className="text-amber-400/80 font-semibold text-sm">note</span>;
+}
@@ -0,0 +1,250 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { CodeBlock } from "./ToolCard";
+
+const MAX_LINE_LENGTH = 200;
+
+const METHOD_COLORS: Record<string, string> = {
+  GET: "text-emerald-400/80", POST: "text-blue-400/80", PUT: "text-yellow-400/80",
+  PATCH: "text-orange-400/80", DELETE: "text-red-400/80",
+};
+
+function statusColor(code: number): string {
+  if (code < 300) return "text-emerald-400/80";
+  if (code < 400) return "text-yellow-400/80";
+  if (code < 500) return "text-orange-400/80";
+  return "text-red-400/80";
+}
+
+/** Hard truncate with trailing "..." */
+function trunc(text: string, maxLen = 80): string {
+  return text.length > maxLen ? text.slice(0, maxLen - 3) + "..." : text;
+}
+
+/** Replace newlines/tabs, then truncate */
+function sanitize(text: string, maxLen = 150): string {
+  return trunc(text.replace(/\n/g, " ").replace(/\r/g, "").replace(/\t/g, " "), maxLen);
+}
+
+/** Limit body to maxLines, each truncated to MAX_LINE_LENGTH-5; returns display string */
+function limitBody(body: string, maxLines: number): string {
+  const lines = body.split("\n");
+  const display = lines.slice(0, maxLines).map(l => trunc(l, MAX_LINE_LENGTH - 5)).join("\n");
+  return lines.length > maxLines ? display + "\n..." : display;
+}
+
+function ListRequests({ args, result }: ToolRendererProps) {
+  const filter = (args.httpql_filter as string) ?? "";
+  const res = result as Record<string, unknown> | null;
+  const rawReqs = res ? res.requests : null;
+  const requests = Array.isArray(rawReqs) ? rawReqs as Array<Record<string, unknown>> : [];
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-purple-400/80 font-semibold text-sm">listing requests</span>
+        {filter && <span className="text-[#888] text-[13px]">{trunc(filter, 150)}</span>}
+      </div>
+      {requests.length > 0 && (
+        <div className="mt-1.5 font-mono text-[13px] space-y-0.5">
+          {requests.slice(0, 20).map((r, i) => {
+            const m = ((r.method as string) ?? "GET").toUpperCase();
+            const host = (r.host as string) ?? "";
+            const path = (r.path as string) ?? "";
+            const resp = r.response as Record<string, unknown> | undefined;
+            const sc = (resp?.statusCode as number) ?? null;
+            return (
+              <div key={i} className="flex gap-2">
+                <span className={`w-10 shrink-0 font-bold ${METHOD_COLORS[m] ?? "text-[#888]"}`}>{m}</span>
+                <span className="text-[#777] truncate">{trunc(host + path, 180)}</span>
+                {sc != null && <span className={`ml-auto shrink-0 ${statusColor(sc)}`}>{sc}</span>}
+              </div>
+            );
+          })}
+          {requests.length > 20 && <div className="text-[#555]">... +{requests.length - 20} more</div>}
+        </div>
+      )}
+    </div>
+  );
+}
+
+function ViewRequest({ args, result }: ToolRendererProps) {
+  const requestId = args.request_id as number | undefined;
+  const part = (args.part as string) ?? "request";
+  const searchPattern = (args.search_pattern as string) ?? "";
+  const res = result as Record<string, unknown> | null;
+  const rawMatches = res ? res.matches : null;
+  const matches = Array.isArray(rawMatches) ? rawMatches as Array<Record<string, string>> : [];
+  const content = res ? (res.content as string) ?? null : null;
+  const hasMore = res ? !!(res.has_more) : false;
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-purple-400/80 font-semibold text-sm">{searchPattern ? "searching" : "viewing"} {part}</span>
+        {requestId != null && <span className="text-[#888] text-[13px]">#{requestId}</span>}
+        {searchPattern && <span className="text-[#666] font-mono text-[13px]">/{trunc(searchPattern, 100)}/</span>}
+      </div>
+      {matches.length > 0 && (
+        <div className="mt-1.5 font-mono text-[13px] space-y-1">
+          {matches.slice(0, 5).map((m, i) => {
+            // Sanitize context: replace newlines with space, trim to 100 chars
+            const before = ((m.before ?? "").replace(/\n/g, " ").replace(/\r/g, "")).slice(-100);
+            const after = ((m.after ?? "").replace(/\n/g, " ").replace(/\r/g, "")).slice(0, 100);
+            return (
+              <div key={i}>
+                {before && <span className="text-[#555]">...{before}</span>}
+                <span className="text-amber-400/80 font-bold">{m.match}</span>
+                {after && <span className="text-[#555]">{after}...</span>}
+              </div>
+            );
+          })}
+          {matches.length > 5 && <div className="text-[#555]">... +{matches.length - 5} more matches</div>}
+        </div>
+      )}
+      {content && !matches.length && (() => {
+        const lines = content.split("\n");
+        const display = lines.slice(0, 15).map(l => trunc(l, MAX_LINE_LENGTH)).join("\n");
+        const showMore = hasMore || lines.length > 15;
+        return (
+          <CodeBlock className="text-[#666]">
+            {display + (showMore ? "\n... more content available" : "")}
+          </CodeBlock>
+        );
+      })()}
+    </div>
+  );
+}
+
+function SendRequest({ args, result }: ToolRendererProps) {
+  const method = ((args.method as string) ?? "GET").toUpperCase();
+  const url = (args.url as string) ?? "";
+  const headers = args.headers as Record<string, string> | undefined;
+  const rawBody = args.body;
+  const reqBody = typeof rawBody === "string" ? rawBody : "";
+  const res = result as Record<string, unknown> | null;
+  const error = res ? (res.error as string) ?? null : null;
+  const statusCode = res ? (res.status_code as number) ?? null : null;
+  const responseTime = res ? (res.response_time_ms as number) ?? null : null;
+  const rawResBody = res ? res.body : null;
+  const resBody = typeof rawResBody === "string" ? rawResBody : null;
+
+  return (
+    <div>
+      <span className="text-purple-400/80 font-semibold text-sm">request</span>
+      <div className="mt-1.5 font-mono text-[13px] space-y-0.5">
+        <div>
+          <span className="text-[#555] select-none mr-1">&gt;&gt;</span>
+          <span className={`font-bold ${METHOD_COLORS[method] ?? "text-[#888]"}`}>{method}</span>
+          <span className="text-[#888] ml-1 break-all">{trunc(url, 180)}</span>
+        </div>
+        {headers && typeof headers === "object" && Object.entries(headers).slice(0, 5).map(([k, v]) => (
+          <div key={k} className="text-[#555] pl-5">{k}: {sanitize(String(v), 150)}</div>
+        ))}
+      </div>
+      {reqBody && (
+        <CodeBlock className="text-[#888]">{limitBody(reqBody, 4)}</CodeBlock>
+      )}
+      {error && <div className="text-red-400/70 text-[13px] mt-1.5">{sanitize(error, 150)}</div>}
+      {statusCode != null && (
+        <div className="font-mono text-[13px] mt-1.5">
+          <span className="text-[#555] select-none mr-1">&lt;&lt;</span>
+          <span className={`font-bold ${statusColor(statusCode)}`}>{statusCode}</span>
+          {responseTime != null && <span className="text-[#555] ml-2">{responseTime}ms</span>}
+        </div>
+      )}
+      {resBody && (
+        <CodeBlock className="text-[#666]">{limitBody(resBody, 6)}</CodeBlock>
+      )}
+    </div>
+  );
+}
+
+function RepeatRequest({ args, result }: ToolRendererProps) {
+  const requestId = args.request_id as number | undefined;
+  const modifications = args.modifications as Record<string, unknown> | undefined;
+  const res = result as Record<string, unknown> | null;
+  const statusCode = res ? (res.status_code as number) ?? null : null;
+  const responseTime = res ? (res.response_time_ms as number) ?? null : null;
+  const rawRepBody = res ? res.body : null;
+  const resBody = typeof rawRepBody === "string" ? rawRepBody : null;
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-purple-400/80 font-semibold text-sm">repeating request</span>
+        {requestId != null && <span className="text-[#888] text-[13px]">#{requestId}</span>}
+      </div>
+      {modifications && typeof modifications === "object" && Object.keys(modifications).length > 0 && (
+        <div className="mt-1.5 font-mono text-[13px] space-y-0.5">
+          {Object.entries(modifications).slice(0, 5).map(([k, v]) => (
+            <div key={k}><span className="text-orange-400/60">{k}:</span> <span className="text-[#777]">{sanitize(typeof v === "string" ? v : JSON.stringify(v), 150)}</span></div>
+          ))}
+        </div>
+      )}
+      {statusCode != null && (
+        <div className="font-mono text-[13px] mt-1.5">
+          <span className="text-[#555] select-none mr-1">&lt;&lt;</span>
+          <span className={`font-bold ${statusColor(statusCode)}`}>{statusCode}</span>
+          {responseTime != null && <span className="text-[#555] ml-2">{responseTime}ms</span>}
+        </div>
+      )}
+      {resBody && (
+        <CodeBlock className="text-[#666]">{limitBody(resBody, 5)}</CodeBlock>
+      )}
+    </div>
+  );
+}
+
+const SCOPE_ACTION: Record<string, string> = {
+  get: "getting", list: "listing", create: "creating", update: "updating", delete: "deleting",
+};
+
+function ScopeRules({ args }: ToolRendererProps) {
+  const action = (args.action as string) ?? "";
+  const scopeName = (args.scope_name as string) ?? "";
+  const label = SCOPE_ACTION[action] ?? (action ? action : "managing");
+  return (
+    <div className="flex items-center gap-2">
+      <span className="text-purple-400/80 font-semibold text-sm">{label} proxy scope</span>
+      {scopeName && <span className="text-[#888] text-[13px]">{trunc(scopeName, 50)}</span>}
+    </div>
+  );
+}
+
+function ListSitemap({ args }: ToolRendererProps) {
+  const parentId = args.parent_id as string | undefined;
+  return (
+    <div className="flex items-center gap-2">
+      <span className="text-purple-400/80 font-semibold text-sm">listing sitemap</span>
+      {parentId && <span className="text-[#888] text-[13px]">under #{trunc(String(parentId), 20)}</span>}
+    </div>
+  );
+}
+
+function ViewSitemapEntry({ args }: ToolRendererProps) {
+  const entryId = args.entry_id as string | undefined;
+  return (
+    <div className="flex items-center gap-2">
+      <span className="text-purple-400/80 font-semibold text-sm">viewing sitemap entry</span>
+      {entryId && <span className="text-[#888] text-[13px]">#{trunc(String(entryId), 20)}</span>}
+    </div>
+  );
+}
+
+export default function ProxyRenderer(props: ToolRendererProps) {
+  switch (props.toolName) {
+    case "list_requests": return <ListRequests {...props} />;
+    case "view_request": return <ViewRequest {...props} />;
+    case "send_request": return <SendRequest {...props} />;
+    case "repeat_request": return <RepeatRequest {...props} />;
+    case "scope_rules": return <ScopeRules {...props} />;
+    case "list_sitemap": return <ListSitemap {...props} />;
+    case "view_sitemap_entry": return <ViewSitemapEntry {...props} />;
+    default:
+      return (
+        <span className="text-purple-400/80 font-semibold text-sm">{props.toolName.replace(/_/g, " ")}</span>
+      );
+  }
+}
@@ -0,0 +1,68 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { CodeBlock, SyntaxBlock } from "./ToolCard";
+
+const MAX_OUTPUT_LINES = 50;
+const MAX_LINE_LENGTH = 200;
+const HEAD = 25;
+const TAIL = 24;
+
+// Full ANSI escape sequence pattern (matches Python's ANSI_PATTERN)
+const ANSI_PATTERN = /\x1b(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~]|\][^\x07]*\x07)/g;
+
+// Strips truncation notices added by Python executor
+const STRIP_PATTERN = /\.\.\. \[(stdout|stderr|result|output|error) truncated at \d+k? chars\]/g;
+
+function stripAnsi(text: string): string {
+  return text.replace(ANSI_PATTERN, "");
+}
+
+function truncateLine(line: string): string {
+  const clean = stripAnsi(line);
+  if (clean.length > MAX_LINE_LENGTH) return clean.slice(0, MAX_LINE_LENGTH - 3) + "...";
+  return clean;
+}
+
+function cleanOutput(output: string): string {
+  return output.replace(STRIP_PATTERN, "").trim();
+}
+
+function formatOutput(output: string): string {
+  const lines = output.split("\n");
+  if (lines.length <= MAX_OUTPUT_LINES) return lines.map(truncateLine).join("\n");
+  const hiddenCount = lines.length - HEAD - TAIL;
+  return [
+    ...lines.slice(0, HEAD).map(truncateLine),
+    `... ${hiddenCount} lines truncated ...`,
+    ...lines.slice(-TAIL).map(truncateLine),
+  ].join("\n");
+}
+
+export default function PythonRenderer({ args, result }: ToolRendererProps) {
+  const action = (args.action as string) ?? "";
+  const code = (args.code as string) ?? (args.script as string) ?? "";
+
+  const res = result as Record<string, unknown> | string | null;
+  let stdout: string | null = null;
+  if (res && typeof res === "object") stdout = typeof res.stdout === "string" ? res.stdout : null;
+  else if (typeof res === "string") stdout = res;
+
+  const subtitle =
+    action === "new_session" ? "new session" :
+    action === "close" ? "close session" :
+    action === "list_sessions" ? "list sessions" : null;
+
+  const output = stdout ? formatOutput(cleanOutput(stdout)) : null;
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-yellow-400/80 font-semibold text-sm">Python</span>
+        {subtitle && <span className="text-[#888] text-[13px]">{subtitle}</span>}
+      </div>
+      {code && <SyntaxBlock code={code} language="python" collapsible />}
+      {output && <CodeBlock className="text-[#666]">{output}</CodeBlock>}
+    </div>
+  );
+}
@@ -0,0 +1,46 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { TruncatedText } from "./ToolCard";
+
+function ScanStartInfo({ args }: ToolRendererProps) {
+  const rawTargets = args.targets;
+  const targets = Array.isArray(rawTargets) ? rawTargets : [];
+  const targetNames = targets.map((t) => (typeof t === "object" && t ? (t.original as string) ?? null : null)).filter(Boolean) as string[];
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-emerald-400/80 font-semibold text-sm">Starting penetration test</span>
+        {targetNames.length === 1 && <span className="text-[#888] text-[13px]">on {targetNames[0]}</span>}
+      </div>
+      {targetNames.length > 1 && (
+        <div className="mt-1.5 space-y-0.5">
+          {targetNames.map((t, i) => (
+            <div key={i} className="text-[13px] text-[#888]"><span className="text-[#555] mr-1">•</span>{t}</div>
+          ))}
+        </div>
+      )}
+    </div>
+  );
+}
+
+function SubagentStartInfo({ args }: ToolRendererProps) {
+  const name = (args.name as string) ?? "Unknown Agent";
+  const task = (args.task as string) ?? "";
+
+  return (
+    <div>
+      <div className="flex items-center gap-2">
+        <span className="text-[#888] text-[13px]">subagent</span>
+        <span className="text-purple-400 font-semibold text-sm">{name}</span>
+      </div>
+      {task && <div className="mt-1.5"><TruncatedText text={task} maxLines={15} /></div>}
+    </div>
+  );
+}
+
+export default function ScanInfoRenderer(props: ToolRendererProps) {
+  if (props.toolName === "subagent_start_info") return <SubagentStartInfo {...props} />;
+  return <ScanStartInfo {...props} />;
+}
@@ -0,0 +1,131 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { CodeBlock, SyntaxBlock } from "./ToolCard";
+
+const MAX_OUTPUT_LINES = 50;
+const MAX_LINE_LENGTH = 200;
+const HEAD = 25;
+const TAIL = 24;
+
+const STRIP_PATTERNS: RegExp[] = [
+  /\n?\[Command still running after [\d.]+s - showing output so far\.?\s*(?:Use C-c to interrupt if needed\.)?\]/g,
+  /^\[Below is the output of the previous command\.\]\n?/gm,
+  /^No command is currently running\. Cannot send input\.$/gm,
+  /^A command is already running\. Use is_input=true to send input to it, or interrupt it first \(e\.g\., with C-c\)\.$/gm,
+];
+
+// Terminal-tool chunk metadata (the OSS engine's shell tool prepends these; the
+// TUI strips them in strix/interface/tui/renderers/shell_renderer.py). Only a
+// contiguous block anchored on a "Chunk ID:" line is stripped, so identical
+// text inside real command output is left untouched.
+const CHUNK_PREAMBLE_START = /^Chunk ID: [0-9a-f]+\s*$/;
+const CHUNK_PREAMBLE_METADATA: RegExp[] = [
+  /^Wall time: [\d.]+ seconds\s*$/,
+  /^Process exited with code -?\d+\s*$/,
+  /^Process running with session ID \d+\s*$/,
+  /^Original token count: \d+\s*$/,
+];
+
+function stripChunkPreambles(lines: string[]): string[] {
+  const out: string[] = [];
+  for (let i = 0; i < lines.length; i++) {
+    if (CHUNK_PREAMBLE_START.test(lines[i])) {
+      let j = i + 1;
+      while (j < lines.length && CHUNK_PREAMBLE_METADATA.some((p) => p.test(lines[j]))) j++;
+      if (j < lines.length && lines[j].trim() === "Output:") j++;
+      i = j - 1;
+      continue;
+    }
+    out.push(lines[i]);
+  }
+  return out;
+}
+
+function escapeRegex(s: string): string {
+  return s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+}
+
+function truncateLine(line: string): string {
+  if (line.length > MAX_LINE_LENGTH) return line.slice(0, MAX_LINE_LENGTH - 3) + "...";
+  return line;
+}
+
+function cleanOutput(raw: string, command: string = ""): string {
+  // Strip ANSI escape sequences and carriage returns
+  let cleaned = raw.replace(/\x1b(?:[@-Z\\-_]|\[[0-?]*[ -/]*[@-~]|\][^\x07]*\x07)/g, "").replace(/\r/g, "");
+
+  for (const pattern of STRIP_PATTERNS) {
+    cleaned = cleaned.replace(pattern, "");
+  }
+
+  if (cleaned.trim()) {
+    const lines = stripChunkPreambles(cleaned.split("\n"));
+    const filtered: string[] = [];
+    for (const line of lines) {
+      // Skip leading blank lines
+      if (filtered.length === 0 && !line.trim()) continue;
+      // Skip [STRIX_N]$ prompt lines
+      if (/^\[STRIX_\d+\]\$\s*/.test(line)) continue;
+      // Skip echoed command (plain)
+      if (command && line.trim() === command.trim()) continue;
+      // Skip echoed command with $/#/> prefix
+      if (command && new RegExp(`^[\\$#>]\\s*${escapeRegex(command.trim())}\\s*$`).test(line)) continue;
+      filtered.push(line);
+    }
+    // Strip trailing [STRIX_N]$ lines
+    while (filtered.length > 0 && /^\[STRIX_\d+\]\$\s*/.test(filtered[filtered.length - 1])) {
+      filtered.pop();
+    }
+    cleaned = filtered.join("\n");
+  }
+
+  return cleaned.trim();
+}
+
+function formatOutput(output: string): string {
+  const lines = output.split("\n");
+  if (lines.length <= MAX_OUTPUT_LINES) return lines.map(truncateLine).join("\n");
+  const hiddenCount = lines.length - HEAD - TAIL;
+  return [
+    ...lines.slice(0, HEAD).map(truncateLine),
+    `... ${hiddenCount} lines truncated ...`,
+    ...lines.slice(-TAIL).map(truncateLine),
+  ].join("\n");
+}
+
+export default function TerminalRenderer({ toolName, args, result }: ToolRendererProps) {
+  const isStdin = toolName === "write_stdin";
+  const command = isStdin
+    ? ((args.chars as string) ?? (args.input as string) ?? "")
+    : ((args.command as string) ?? (args.cmd as string) ?? "");
+
+  const res = result as Record<string, unknown> | string | null;
+  let content: string | null = null;
+  let error: string | null = null;
+  let exitCode: number | null = null;
+
+  if (res && typeof res === "object") {
+    content = typeof res.content === "string" ? res.content : null;
+    error = typeof res.error === "string" ? res.error : null;
+    exitCode = typeof res.exit_code === "number" ? res.exit_code : null;
+    const s = typeof res.status === "string" ? res.status : "";
+    if (s === "running" || s === "command still running") content = null;
+  } else if (typeof res === "string") {
+    content = res;
+  }
+
+  const output = content ? formatOutput(cleanOutput(content, command)) : null;
+
+  return (
+    <div>
+      <span className="text-emerald-400/80 font-semibold text-sm">{isStdin ? "Terminal input" : "Terminal"}</span>
+      {command && <SyntaxBlock code={command} language="bash" collapsible />}
+      {error && <CodeBlock className="text-red-400/70">{error}</CodeBlock>}
+      {output && <CodeBlock className="text-[#666]">{output}</CodeBlock>}
+      {exitCode != null && exitCode !== 0 && (
+        <div className="font-mono text-[13px] text-red-400/70 mt-0.5">exit code {exitCode}</div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,18 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { TruncatedText } from "./ToolCard";
+
+export default function ThinkRenderer({ args }: ToolRendererProps) {
+  const thought = (args.thought as string) ?? (args.content as string) ?? "";
+  if (!thought) return null;
+
+  return (
+    <div>
+      <span className="text-purple-400/80 font-semibold text-sm">Agent is thinking</span>
+      <div className="mt-1.5 italic text-[#888]">
+        <TruncatedText text={thought} maxLines={20} />
+      </div>
+    </div>
+  );
+}
@@ -0,0 +1,120 @@
+"use client";
+
+import type { ToolRendererProps } from "@/types/events";
+import { Circle, CircleDot, CircleCheckBig, Trash2, Plus, RefreshCw, CheckCheck, RotateCcw, Pencil } from "lucide-react";
+
+interface TodoItem {
+  id?: string;
+  title?: string;
+  status?: string;
+}
+
+const ACTION_LABELS: Record<string, { label: string; Icon: typeof Circle }> = {
+  create_todo: { label: "Task added", Icon: Plus },
+  list_todos: { label: "Plan", Icon: CheckCheck },
+  update_todo: { label: "Task updated", Icon: Pencil },
+  mark_todo_done: { label: "Task completed", Icon: CircleCheckBig },
+  mark_todo_pending: { label: "Task reopened", Icon: RotateCcw },
+  delete_todo: { label: "Task removed", Icon: Trash2 },
+};
+
+function StatusIcon({ status }: { status: string }) {
+  if (status === "done") return <CircleCheckBig className="w-3.5 h-3.5 text-emerald-400/70 shrink-0" />;
+  if (status === "in_progress") return <CircleDot className="w-3.5 h-3.5 text-purple-400/70 shrink-0 animate-pulse" />;
+  return <Circle className="w-3.5 h-3.5 text-[#444] shrink-0" />;
+}
+
+function TodoList({ todos, highlightId }: { todos: TodoItem[]; highlightId?: string }) {
+  return (
+    <div className="space-y-0">
+      {todos.map((todo, i) => {
+        const s = todo.status ?? "pending";
+        const isHighlighted = highlightId && todo.id === highlightId;
+        return (
+          <div
+            key={todo.id ?? i}
+            className={`flex items-start gap-2.5 py-1.5 px-2 -mx-2 rounded-md transition-colors ${
+              isHighlighted ? "bg-purple-500/[0.08]" : ""
+            }`}
+          >
+            <div className="mt-[1px]">
+              <StatusIcon status={s} />
+            </div>
+            <span
+              className={`text-[13px] leading-snug ${
+                s === "done"
+                  ? "text-[#555] line-through"
+                  : s === "in_progress"
+                    ? "text-[#bbb]"
+                    : "text-[#999]"
+              }`}
+            >
+              {todo.title ?? "(untitled)"}
+            </span>
+          </div>
+        );
+      })}
+    </div>
+  );
+}
+
+export default function TodoRenderer({ toolName, args, result }: ToolRendererProps) {
+  const action = ACTION_LABELS[toolName] ?? { label: "Plan", Icon: RefreshCw };
+  const ActionIcon = action.Icon;
+  const res = result as Record<string, unknown> | string | null;
+
+  // Simple string result
+  if (typeof res === "string" && res.trim()) {
+    return (
+      <div>
+        <div className="flex items-center gap-2">
+          <ActionIcon className="w-3.5 h-3.5 text-purple-400/60" />
+          <span className="text-purple-400/80 font-semibold text-sm">{action.label}</span>
+        </div>
+        <div className="mt-1.5 text-[#888] text-[13px]">{res.trim()}</div>
+      </div>
+    );
+  }
+
+  // Parse structured result
+  let todos: TodoItem[] = [];
+  let error: string | null = null;
+  let todoId: string | undefined;
+
+  if (res && typeof res === "object") {
+    error = (res.error as string) ?? null;
+    if (res.success) {
+      const rawTodos = res.todos;
+      todos = Array.isArray(rawTodos) ? (rawTodos as TodoItem[]) : [];
+    }
+    todoId = (res.id as string) ?? (args.todo_id as string) ?? undefined;
+  }
+
+  // For mutations, highlight the affected item
+  const highlightId = toolName !== "list_todos" ? todoId : undefined;
+
+  // No todos and no error — brief label only
+  if (todos.length === 0 && !error) {
+    return (
+      <div className="flex items-center gap-2">
+        <ActionIcon className="w-3.5 h-3.5 text-purple-400/60" />
+        <span className="text-purple-400/80 font-semibold text-sm">{action.label}</span>
+      </div>
+    );
+  }
+
+  return (
+    <div>
+      <div className="flex items-center gap-2 mb-2">
+        <ActionIcon className="w-3.5 h-3.5 text-purple-400/60" />
+        <span className="text-purple-400/80 font-semibold text-sm">{action.label}</span>
+      </div>
+      {error && <div className="text-red-400/70 text-[13px] mb-2">{error}</div>}
+      {todos.length > 0 && (
+        <div className="rounded-lg border border-white/[0.06] bg-white/[0.015] px-3 py-2">
+          <TodoList todos={todos} highlightId={highlightId} />
+        </div>
+      )}
+    </div>
+  );
+}
@@ -0,0 +1,100 @@
+"use client";
+
+import { useState } from "react";
+import Markdown from "./Markdown";
+import hljs from "@/lib/hljs";
+import "highlight.js/styles/github-dark.css";
+
+const OUTPUT_PREVIEW_LINES = 6;
+const CODE_PREVIEW_LINES = 20;
+
+/** Truncatable markdown text with "Show more" */
+export function TruncatedText({ text, maxLines = 20 }: { text: string; maxLines?: number }) {
+  const [expanded, setExpanded] = useState(false);
+  const lines = text.trimEnd().split("\n");
+  const needsTruncation = lines.length > maxLines;
+
+  return (
+    <div>
+      <div
+        className={expanded && needsTruncation ? "max-h-[1200px] overflow-auto" : ""}
+        style={!expanded && needsTruncation ? { display: "-webkit-box", WebkitLineClamp: maxLines, WebkitBoxOrient: "vertical", overflow: "hidden" } : undefined}
+      >
+        <Markdown text={text} />
+      </div>
+      {needsTruncation && (
+        <button onClick={() => setExpanded(!expanded)} className="text-xs text-[#555] hover:text-[#888] mt-1">
+          {expanded ? "Show less" : "Show more"}
+        </button>
+      )}
+    </div>
+  );
+}
+
+/** Code/output block — truncates to 12 lines with "Show more", expanded view scrolls */
+export function CodeBlock({ children, className = "" }: { children: React.ReactNode; className?: string }) {
+  const [expanded, setExpanded] = useState(false);
+
+  const isString = typeof children === "string";
+  const lines = isString ? (children as string).trimEnd().split("\n") : null;
+  const needsTruncation = lines !== null && lines.length > OUTPUT_PREVIEW_LINES;
+  const displayContent = needsTruncation && !expanded
+    ? lines!.slice(0, OUTPUT_PREVIEW_LINES).join("\n")
+    : children;
+
+  return (
+    <div>
+      <pre className={`font-mono text-[13px] leading-relaxed whitespace-pre-wrap break-words mt-1 ${
+        expanded ? "overflow-auto max-h-[1200px]" : "overflow-hidden"
+      } ${className}`}>
+        {displayContent}
+      </pre>
+      {needsTruncation && (
+        <button
+          onClick={() => setExpanded(!expanded)}
+          className="text-xs text-[#555] hover:text-[#888] mt-0.5"
+        >
+          {expanded ? "Show less" : "Show more"}
+        </button>
+      )}
+    </div>
+  );
+}
+
+/** Syntax-highlighted code block — no border, no line numbers, just highlighting.
+ *  Pass `collapsible` to get a "Show more" toggle instead of a scroll cap. */
+export function SyntaxBlock({ code, language, className = "", collapsible = false }: { code: string; language?: string; className?: string; collapsible?: boolean }) {
+  const [expanded, setExpanded] = useState(false);
+
+  const lines = code.trimEnd().split("\n");
+  const needsTruncation = collapsible && lines.length > CODE_PREVIEW_LINES;
+  const displayCode = needsTruncation && !expanded
+    ? lines.slice(0, CODE_PREVIEW_LINES).join("\n")
+    : code;
+
+  let highlighted: string;
+  try {
+    highlighted = language
+      ? hljs.highlight(displayCode, { language, ignoreIllegals: true }).value
+      : hljs.highlightAuto(displayCode).value;
+  } catch {
+    highlighted = hljs.highlightAuto(displayCode).value;
+  }
+
+  return (
+    <div>
+      <pre className={`font-mono text-[12px] leading-relaxed px-0 py-1 mt-1 whitespace-pre-wrap break-all ${
+        collapsible
+          ? expanded ? "overflow-auto max-h-[1200px]" : "overflow-hidden"
+          : "overflow-auto max-h-[400px]"
+      } ${className}`}>
+        <code dangerouslySetInnerHTML={{ __html: highlighted }} />
+      </pre>
+      {needsTruncation && (
+        <button onClick={() => setExpanded(!expanded)} className="text-xs text-[#555] hover:text-[#888] mt-0.5">
+          {expanded ? "Show less" : "Show more"}
+        </button>
+      )}
+    </div>
+  );
+}
--- a/Show More
+++ b/Show More