rag-patterns
Pass
Audited by Gen Agent Trust Hub on Jun 23, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides architectural best practices and educational content for RAG implementations, including hybrid search and semantic chunking.
- [COMMAND_EXECUTION]: The documentation includes example shell commands (e.g., 'make index', 'python scripts/evaluate_rag.py') intended for developer-side system setup and evaluation. These are not instructions for the agent to execute malicious commands.
- [PROMPT_INJECTION]: The retrieval patterns described (CRAG, Multi-hop) ingest external data from a knowledge base, presenting a standard Indirect Prompt Injection surface. 1. Ingestion points: 'hybrid_search_kb', 'smart_query', 'crag_search', 'multi_hop_search' tools. 2. Boundary markers: None explicitly shown in snippets. 3. Capability inventory: Access is strictly limited to the 'Read' tool. 4. Sanitization: Not specified in the logic. This surface is inherent to RAG functionality and is appropriately scoped via the 'allowed-tools' restriction.
Audit Metadata