babysitting-pr

Pass

Audited by Gen Agent Trust Hub on Apr 12, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by consuming untrusted data from external sources and acting upon it.\n
  • Ingestion points: The agent fetches and processes GitHub pull request comments via the GitHub API and CI failure logs via gh run view (SKILL.md).\n
  • Capability inventory: The agent is empowered to modify project files, execute local build and linting scripts, and push these changes back to the remote repository.\n
  • Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its primary goals and potentially malicious instructions embedded in the external data it reads.\n
  • Sanitization: No content validation or sanitization is performed on the ingested comments or logs before the agent attempts to implement described "fixes."\n- [COMMAND_EXECUTION]: The skill executes local scripts which can perform arbitrary actions based on the project's own configuration.\n
  • Evidence: The skill instructs the agent to run npm run lint, npm run build, and npx tsc (SKILL.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 12, 2026, 09:43 PM