source-driven-development
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches external documentation, which constitutes an ingestion point for untrusted data. The instructions provide explicit mitigation strategies: treating fetched content strictly as data, using clear extraction boundaries (API definitions and usage examples only), and explicitly ignoring instructions within the content that attempt to redirect the agent's behavior (e.g., 'ignore previous instructions').
- [SAFE]: The skill reads local dependency files such as package.json, requirements.txt, and go.mod. This behavior is restricted to tech stack and version detection, which is standard for development-oriented tools and does not involve accessing sensitive user data or credentials.
- [SAFE]: The skill includes explicit warnings against executing commands or hardcoding external endpoints discovered in documentation, significantly reducing the risk of remote code execution or accidental data exfiltration triggers found in untrusted documentation examples.
- [PROMPT_INJECTION]: While detectors flagged phrases like 'ignore previous instructions', these appear exclusively within a defensive context, instructing the agent on how to identify and disregard malicious content within external documentation rather than attempting to override the agent's core safety guidelines.
Audit Metadata