apply-findings
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process 'findings' which are external data points often derived from untrusted sources like third-party code or pull request descriptions. This creates a surface for indirect prompt injection if the findings themselves contain malicious instructions designed to trick the agent during the application phase.
- Ingestion points: The conversation context containing evaluation results (Step 1).
- Boundary markers: The skill requires findings to have been pre-processed by
/evaluate-findings, establishing a procedural boundary. - Capability inventory: The agent is authorized to modify files (Step 2, Item 7).
- Sanitization: The skill includes rigorous manual verification logic in Step 2, Item 3, which instructs the agent to 'construct an input that clears the point the fix guards', 'trace whether it still reaches that failure', and 'verify it independently' rather than relying on the proposer's authority. These instructions function as a logical filter against malicious or unsound code suggestions.
Audit Metadata