The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for implementing RAG systems that ingest and process untrusted data from external sources such as vector databases and document stores. This creates a surface for indirect prompt injection, where malicious instructions hidden within retrieved documents could influence the agent's behavior.
Ingestion points: Data enters the context via PineconeVectorStore, WeaviateVectorStore, Chroma, and PGVector retrievers in SKILL.md and references/details.md.
Boundary markers: The prompt templates in SKILL.md and references/details.md utilize structural delimiters and instructions such as "Answer based on the context below" and "If you cannot answer... say so" to help isolate external data.
Capability inventory: The skill utilizes LLMs to generate responses and execute retrieval logic using standard LangChain and LangGraph components.
Sanitization: The provided code does not implement explicit sanitization or content filtering for the retrieved document content before it is interpolated into LLM prompts.

rag-implementation