rag
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill functions as an educational resource for RAG development and contains no malicious directives or executable threats.
- [DATA_EXFILTRATION]: Technical recommendations include the use of tenant-based filtering in vector and keyword searches, supporting data isolation best practices to prevent unauthorized data exposure.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies and provides mitigations for potential injection surfaces where external data is processed.
- Ingestion points: The GROUNDED_PROMPT template in SKILL.md interpolates retrieved passages and user questions.
- Boundary markers: The template employs clear constraints, instructing the agent to use only provided context and specifying a mandatory citation format.
- Capability inventory: The examples in SKILL.md describe retrieval operations (vector_store.search, bm25.search) and reranking processes.
- Sanitization: The workflow incorporates a refusal mechanism, directing the agent to respond with a specific string if the context is insufficient.
Audit Metadata