full-babysit-pr

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources.
  • Ingestion points: Processes pull request comments and CI failure logs to analyze and implement fixes (SKILL.md, references/pr-comments.md).
  • Boundary markers: Does not specify explicit delimiters for untrusted data, but manages responses with 'Confirmed' and 'Automatic' prefixes to distinguish between user-approved and agent-generated content.
  • Capability inventory: Possesses capabilities to modify local files, commit changes, push to remote branches, and perform merges (SKILL.md).
  • Sanitization: Enforces a mandatory manual user approval step before any code modifications are published or pushed to the remote repository. This human gate acts as the primary defense against indirect prompt injection from malicious PR comments.
  • [COMMAND_EXECUTION]: The skill runs local repository verification tools.
  • Evidence: Executes testing and linting commands (e.g., vitest, jest) as defined in the host repository's own instructions (references/simplification.md, references/verification.md). This execution is part of the tool's intended PR maintenance functionality.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 01:12 AM