address-pr-comments
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from pull request comments and discussion threads, which could be used to deliver malicious instructions.\n- Ingestion points: External pull request comments and reply threads (SKILL.md).\n- Boundary markers: The skill requires verification of the comment's premise against the actual source code and existing reply threads.\n- Capability inventory: The skill is capable of modifying the local filesystem to apply code fixes and resolves PR threads via tool calls.\n- Sanitization: The primary defense is a mandatory human-in-the-loop confirmation ('Per-fix flow') for every individual code modification.
Audit Metadata