The Agent Skills Directory

[SAFE]: The skill serves as a reference guide for LLM application patterns. All code snippets provided are educational examples of standard industry practices such as RAG (Retrieval-Augmented Generation), agent architectures (ReAct, Plan-and-Execute), and LLMOps monitoring.
[SAFE]: External references target well-known and trusted repositories/platforms (Dify, LangChain, LlamaIndex, Anthropic Cookbook). These are documented neutrally as resources.
[INDIRECT_PROMPT_INJECTION]: The skill describes patterns that ingest untrusted user data (e.g., in RAG prompts and Agent tools). While these represent an attack surface, the skill includes best practices such as instruction delimiters and structured data handling to mitigate these risks. As these are educational patterns and not a deployed system, the severity is safe.
[COMMAND_EXECUTION]: Code examples include calls to external tools and functions. These are standard architectural patterns for agents and do not contain hardcoded malicious commands or unauthorized file access.

llm-app-patterns