babysitting-pr
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources that could contain malicious instructions designed to manipulate agent behavior.\n
- Ingestion points: Steps 3 and 4 read CI failure logs (
gh run view --log-failed) and PR review comments (gh api .../comments).\n - Boundary markers: Absent. The instructions do not specify delimiters or warnings to ignore instructions embedded in the logs or comments.\n
- Capability inventory: The skill has significant capabilities including
git push,npm run build,npm run lint, and arbitrary file modification viagit commit.\n - Sanitization: Absent. The skill analyzes external content directly to determine and apply code fixes.\n- [COMMAND_EXECUTION]: The skill executes local scripts such as
npm run lintandnpm run build. While these are standard development workflows, they involve executing code defined in the repository'spackage.jsonfile, which could perform unauthorized actions if the repository contains malicious configurations.
Audit Metadata