llm-app-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides architectural patterns for Retrieval-Augmented Generation (RAG) and AI agents that ingest external data.
- Ingestion points: External document content is interpolated into the
RAG_PROMPT_TEMPLATEwithinSKILL.md. - Boundary markers: The prompt template includes a restrictive instruction to "Answer the user's question based ONLY on the following context," which serves as a delimiter for the model.
- Capability inventory: The
ReActAgentandFunctionCallingAgentpatterns include functions for executing tools (_execute_tool) based on model output. - Sanitization: The conceptual Python code snippets provided do not explicitly demonstrate input sanitization or validation of the retrieved content.
- [EXTERNAL_DOWNLOADS]: The skill includes links to several well-known and trusted industry resources.
- References the Dify platform on GitHub.
- Links to official documentation for LangChain and LlamaIndex.
- References the Anthropic Cookbook repository.
Audit Metadata