file-pr
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill incorporates an explicit approval boundary in Step 3 of SKILL.md, requiring the agent to stop and obtain user authorization before performing external communications such as filing a pull request.
- [SAFE]: Guidelines in Step 2 of SKILL.md specifically instruct the agent not to include secrets, private logs, or unsupported claims in the pull request body, reducing the risk of accidental data exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from repository issues and specifications which could contain instructions designed to influence the agent. 1. Ingestion points: Issue descriptions, specifications, and diffs mentioned in Step 1 of SKILL.md. 2. Boundary markers: A mandatory 'Approval boundary' defined in Step 3 of SKILL.md requires human review of all generated content. 3. Capability inventory: Drafting and filing pull requests as described in Step 4 of SKILL.md. 4. Sanitization: Human-in-the-loop validation is enforced as the primary mitigation for processing external content.
Audit Metadata