The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements RAG patterns that ingest external data from retrievers and web searches, which could contain malicious instructions designed to influence the agent's behavior. * Ingestion points: Untrusted data enters the context through document retrieval and web search results in the CorrectiveRAG and SelfRAG classes in SKILL.md. * Boundary markers: The prompts in SKILL.md (e.g., _refine_knowledge and _generate) do not use robust delimiters or instructions to ignore commands embedded within the retrieved documents. * Capability inventory: The skill incorporates network search capabilities via the TavilySearchResults tool. * Sanitization: The skill lacks logic to sanitize or escape retrieved content before interpolating it into LLM prompts.

corrective-rag