ai-features
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill effectively mitigates indirect prompt injection risks by teaching developers to isolate untrusted content. Instruction 6 in SKILL.md and Instruction 3 in RAG.md command the agent to treat retrieved documents and user files strictly as data, ensuring that any embedded instructions (like 'ignore previous instructions') are quoted or ignored rather than executed.
- [SAFE]: Code examples utilize best practices for credential management, using environment variables for API keys instead of hardcoding. Interaction with external services like AssemblyAI is documented through official API patterns.
- [SAFE]: The extraction logic in EXTRACTION.md leverages structured output validation using the Zod library, ensuring that AI-generated content conforms to expected schemas and undergoes domain-specific verification.
- [SAFE]: The retrieval patterns in RAG.md and VECTOR-SEARCH.md emphasize the importance of applying permissions and Row Level Security (RLS) before data is processed by the AI, preventing cross-tenant data leakage.
Audit Metadata