llm-app-patterns
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides architectures for Retrieval-Augmented Generation (RAG) and AI Agents that process external, untrusted content from documents or tool outputs.
- Ingestion points: External data enters the context during document ingestion and tool observation phases (SKILL.md).
- Boundary markers: Templates include basic instructions to limit model focus (e.g., "Answer... based ONLY on context"), though these do not fully prevent adversarial injection.
- Capability inventory: The agent patterns include tool execution (execute_tool) and external search capabilities.
- Sanitization: The provided architectural snippets do not include explicit input sanitization or validation logic.
Audit Metadata