The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents architectures for RAG (Retrieval-Augmented Generation) and autonomous agents that process external, potentially untrusted data.
Ingestion points: Functions like hybrid_search and compressed_retrieval in SKILL.md retrieve external content for processing.
Boundary markers: The provided RAG_PROMPT_TEMPLATE uses explicit instructions ('Answer... based ONLY on the following context') to constrain agent behavior.
Capability inventory: The agent patterns (FunctionCallingAgent, ReActAgent) describe the use of external tools such as search_web and calculate.
Sanitization: The reference implementation uses standard Python string interpolation for prompt building without specialized sanitization or escaping of external inputs.

llm-app-patterns