fix(context): clamp shell output cap and count byte-trimmed dropped lines

Treat tool_output_max_tokens as a ceiling so an explicit model-supplied
cap can't exceed it, and derive the truncation notice's dropped-line
count from the lines actually kept after the byte-trim pass. Also cast
the pygments fallback lexer so it satisfies the resolve_lexer return
type under the pre-commit mypy hook.

tests/test_agent_factory_shell.py

@@ -41,7 +41,7 @@ async def test_wrap_exec_command_defaults_shell_to_bash() -> None:
 
 
 @pytest.mark.asyncio
-async def test_wrap_exec_command_preserves_explicit_output_cap() -> None:
+async def test_wrap_exec_command_preserves_smaller_explicit_output_cap() -> None:
     captured: dict[str, str] = {}
     wrapped = factory._wrap_exec_command(_capturing_exec_tool(captured))
 
@@ -52,6 +52,20 @@ async def test_wrap_exec_command_preserves_explicit_output_cap() -> None:
     assert json.loads(captured["raw_input"])["max_output_tokens"] == 42
 
 
+@pytest.mark.asyncio
+async def test_wrap_exec_command_clamps_oversized_explicit_output_cap() -> None:
+    captured: dict[str, str] = {}
+    wrapped = factory._wrap_exec_command(_capturing_exec_tool(captured))
+    ceiling = load_settings().context.tool_output_max_tokens
+
+    await wrapped.on_invoke_tool(
+        cast("Any", None),
+        json.dumps({"cmd": "echo hi", "max_output_tokens": ceiling * 100}),
+    )
+
+    assert json.loads(captured["raw_input"])["max_output_tokens"] == ceiling
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("shell", ["/bin/zsh", ""])
 async def test_wrap_exec_command_preserves_explicit_shell(shell: str) -> None:

tests/test_output_store.py

@@ -2,6 +2,8 @@
 
 from __future__ import annotations
 
+import re
+
 from strix.tools.output_store import bound_text
 
 
@@ -44,3 +46,18 @@ def test_notice_reports_dropped_counts() -> None:
 
     assert "lines" in bounded
     assert "bytes" in bounded
+
+
+def test_dropped_line_count_accounts_for_byte_trimming() -> None:
+    # A tight byte budget forces the byte pass to drop whole lines from the
+    # head/tail slices; the notice must count those, not just the middle.
+    text = "\n".join(f"line-{i}" for i in range(200))
+    bounded = bound_text(text, max_lines=20, max_bytes=40)
+
+    match = re.search(r"\[\.\.\. (\d+) lines", bounded)
+    assert match is not None, bounded
+    dropped = int(match.group(1))
+    kept = [ln for ln in bounded.splitlines() if ln and "truncated" not in ln]
+    assert dropped == 200 - len(kept)
+    # The naive middle-only count (max_lines split evenly) would under-report.
+    assert dropped > 200 - 20