fix-all
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external pull request reviews and discussions.
- Ingestion points: Pull request comments, CodeRabbit review threads, and git diffs are ingested into the agent context.
- Boundary markers: The skill includes a dedicated 'Security model' section that explicitly identifies these inputs as adversarial data and instructs the agent to never treat them as instructions.
- Capability inventory: The skill possesses the ability to perform git operations (commit, push) and modify PR metadata via
gh pr edit. - Sanitization: Safety is maintained through explicit hard prohibitions in the prompt and a corresponding platform-level permission allowlist in
.claude/settings.json.
Audit Metadata