#173 [ROSETTA] dangerous-actions hook: MCP cmd/shell_command fields can't carry override marker

[isolomatov-gd]

Closes #173

Summary

• Derive MCP marker-eligible fields from the shell and content field sets.
• Add marked and unmarked cmd and shell_command regression coverage.
• Assert that a marker in an MCP path cannot override a dangerous command.

Testing

• npm --prefix src/hooks run check
• npm --prefix src/hooks exec vitest run tests/dangerous-actions.test.ts (201 passed)
• python3 scripts/pre_commit.py with a writable temporary npm cache (35 files, 1,446 hook tests passed)

Assumptions

• MCP_PATH_FIELDS remains intentionally excluded from marker eligibility, as specified in the approved plan.
• The extracted runner has no root Python virtual environment, so repository Python checks were skipped by the existing validation script; hook validation completed.

[github-actions]

Rosetta Triage Review

Summary: This PR fixes the MCP override-marker gap for cmd and shell_command by deriving marker-eligible fields from the shell and content field sets. It stays focused on issue #173 and preserves the deliberate exclusion of path fields.

Findings:

• No blocking issue found.
• Regression coverage verifies marked and unmarked cmd and shell_command inputs, plus that a path-only marker cannot override a dangerous command.

Validation:

• Available build/test, static-analysis, and automated security checks pass.

Automated triage by Rosetta agent