The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill outlines architectural patterns for Retrieval-Augmented Generation (RAG) and AI Agents that involve ingesting data from external, potentially untrusted sources into prompt contexts. This architecture presents a surface for indirect prompt injection attacks.
Ingestion points: External data is processed and interpolated into LLM prompts through variables such as {context} in the RAG pipeline and Observation results in the ReAct agent loop.
Boundary markers: The provided code snippets use basic labels (e.g., "Context:", "Question:") to separate data, which is a standard but minimal boundary practice.
Capability inventory: The patterns describe agents capable of performing web searches and executing mathematical calculations.
Sanitization: As these are educational examples, the snippets do not include explicit input sanitization or safety-filtering logic.
[EXTERNAL_DOWNLOADS]: The skill contains references and links to well-known industry resources and frameworks, including LangChain, LlamaIndex, Dify, and the Anthropic Cookbook. These links point to official GitHub repositories and documentation for developer reference.

llm-app-patterns