---
title: "Action Safety"
description: "Review potentially dangerous actions before they execute"
---

Action safety is enabled by default and is independent of scan depth. `quick`,
`standard`, and `deep` control coverage; guarded review controls which effects
may be executed.

```bash
strix --target https://example.test
```

Guarded review permits non-destructive interaction after contextual review,
including injection probes, reconnaissance, enumeration, and fuzzing. Actions
judged destructive or persistent are blocked.

## Disabling Safety

Use the explicit dangerous opt-out only when external containment makes it
necessary:

```bash
strix --target https://example.test --dangerously-disable-safety
```

This disables both action review and workspace isolation. Local directories are
mounted live and writable. A run created with safety disabled requires the flag
again when resumed; a guarded run cannot be downgraded while resuming.

## Contextual Review

Before an ambiguous shell or browser action executes, Strix compiles a frozen
evidence packet containing the effective command, target scope, relevant script
source and imports, prior tool-call evidence, browser snapshot context, and
workspace persistence details.

The safety model may decide immediately or make exactly one `run_inspection`
tool call. That call runs a Python standard-library analysis script in a
separate networkless, read-only container over the frozen evidence. If the tool
is used, the model's next response must be the final decision.

The review is bounded to at most two model turns and one optional inspection
call. Timeouts, malformed decisions, a second tool call, incomplete evidence,
and reviewer failures fail closed.

In the interactive TUI, the reviewer can defer when complete evidence still
leaves genuine ambiguity about whether an exact action is dangerous. Strix then
pauses that tool call and asks the user to approve or deny it. Denial is selected
by default, Escape denies, and the request waits until it is answered, the agent
is stopped, or Strix exits. Approval applies only to the frozen call shown in
the prompt; actions too large to display exactly must be split into smaller
tool calls. Deterministic blocks, incomplete evidence, review errors, and
actions confidently judged dangerous cannot be overridden.

Non-interactive runs have no human approval channel. Ambiguity and
low-confidence decisions continue to block, preserving fail-closed autonomous
behavior.

The reviewer judges an action by its effect, not by the technique it uses or by
whether a hostname appears in target scope. A read-only injection probe (a boolean,
`UNION SELECT`, or time-based payload), a reflected-input test, or recon passes;
a payload that writes or destroys (`DROP`, `DELETE`, `INSERT`, `INTO OUTFILE`,
stacked statements, command execution), a mutating request, or any persistent
change is blocked or, in the TUI, deferred when its effect is genuinely ambiguous.

Scope still controls what Strix actively tests, but the safety reviewer is not a
scope enforcement layer. Ordinary passive requests to research services such as
`crt.sh`, DNS and WHOIS, package registries, search, and public documentation are
allowed when they support an authorized target. Those services do not become
targets for scanning or exploitation.

## Deterministic Rules

Some outcomes never reach the model. Destructive commands, environment
overrides that change which code an interpreter loads (`PYTHONPATH`,
`LD_PRELOAD`, `AGENT_BROWSER_SESSION`, and similar), and blocked browser actions
are refused outright. A small set of
read-only commands is allowed outright, but only when its options are also
read-only: `rg --pre` and anything else that hands the command another program
to run goes to review instead.

Browser observation commands are allowed outright only in the form that just
reads: `tab` lists tabs, but `tab new <url>` navigates and `tab close` discards
page state, so a grouped verb with a subcommand goes to review.

Commands that wrap another program (`sudo`, `timeout`, `xargs`, `nohup`, and
similar) and interactive `write_stdin` payloads cannot be resolved to a single
effective action before dispatch, so they are blocked. Issue the command as its
own `exec_command` call.

## Scripts

When a command executes a script, Strix reads the current entrypoint and local
Python imports without importing or running them. Inline `python -c` source is
analyzed the same way. Absolute imports resolve against the entrypoint's
directory and relative imports against the importing module's package, and an
imported name is followed as a submodule as well as an attribute, so the whole
local closure is inspected. Decisions bind to content hashes. Dynamic code
execution, import-path mutation, unresolved generated commands, oversized
dependency closures, entrypoints outside `/workspace`, and unsupported evidence
block the action.

A command that runs code Strix cannot resolve to an inspectable script — an
unrecognized interpreter, or an interpreter given no script — is blocked rather
than reviewed against an empty evidence packet.

When a command reads a workspace data file — through input redirection
(`while read … done < hosts.txt`) or a target-list flag (`ffuf -w words.txt`,
`httpx -l hosts.txt`) — that file's contents are attached to the packet so the
reviewer can assess the exact entries, queried hosts, or fuzz inputs instead of
blocking because it cannot see them. Only workspace-resident files are read; an
oversize file is attached truncated. Any workspace change while the action is
under review or awaiting approval invalidates the decision.

Browser automation inside scripts is blocked in safety modes. Issue browser
operations as individual raw `agent-browser` commands so each action can be
reviewed against the current snapshot and element references.

Commands that create and execute code in one shell expression should be split
into separate creation and execution calls.

## Browser Commands

Strix continues to use the raw `agent-browser` CLI. In safety modes it assigns
an isolated browser session per agent and rejects model-supplied session,
profile, or CDP overrides.

Interactions with element references require a prior recorded snapshot. A
snapshot taken before a navigation or any other page-changing action is stale:
the action is blocked and the agent must snapshot again.

Composite operations such as `auth login`, arbitrary `eval`, browser state
persistence, and uploads are blocked. Guarded login should use explicit fill
and submit steps with credentials supplied in the initial user instruction.

## Workspace Isolation

By default, user-owned local directories are copied into:

```text
strix_runs/<run>/.state/workspaces/<name>
```

The copy is mounted writable, while the original source remains unchanged.
`.git`, `.agents`, and `.codex` inside the copy stay read-only: they carry
repository and agent-instruction state that survives `--resume`. Copies are
retained for resume. Repository targets are already cloned into a disposable
location and do not need another copy.

In-tree symlinks are materialized. Dangling, cyclic, device, and out-of-tree
symlinks are omitted. Files are copied rather than hard-linked.

## Limitations

Contextual review reduces accidental harmful actions; it is not a complete
network containment boundary. Arbitrary dynamic programs, raw sockets, or
processes that ignore proxy settings cannot always be predicted statically.
Unresolvable behavior blocks in safety modes.

Deterministic rules cover the cases listed above. Every other command is judged
by the safety model against compiled evidence, so a tool whose effects are not
statically recognizable — a scanner or exploit framework that mutates the
target through its own protocol, for example — rests on that judgment rather
than on a rule. Strong containment additionally requires externally enforced
egress policy and reduced sandbox privileges.