The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for Retrieval-Augmented Generation (RAG), which involves ingesting and processing external document data that could potentially contain adversarial instructions.
Ingestion points: External document content is ingested via retriever.ainvoke and various document loaders/splitters in SKILL.md and references/details.md.
Boundary markers: The code samples include prompt templates that use labels like Context: to delimit retrieved data and explicit instructions to stay within the provided context.
Capability inventory: The logic is limited to LLM calls for generation and metadata retrieval; it does not include risky capabilities like shell execution or file system modification.
Sanitization: The examples rely on prompt engineering constraints to handle untrusted context data.

rag-implementation