The Agent Skills Directory

[SAFE]: The skill content is educational, focusing on LangChain and LangGraph architecture with provided code examples that align with industry best practices.
[INDIRECT_PROMPT_INJECTION]: The architectural patterns for RAG (Retrieval-Augmented Generation) and multi-agent workflows involve ingesting external data into the LLM context (e.g., in the search_database tool and retrieve node in SKILL.md). This creates an attack surface for indirect prompt injection. However, the skill provides examples that include boundary markers (e.g., "Answer based on the context below") and does not implement exploitable execution chains.
[DYNAMIC_EXECUTION]: The provided calculate tool uses Python's ast module to safely parse and evaluate mathematical expressions by traversing the abstract syntax tree and only allowing specific, safe operators. This is a secure alternative to using the dangerous eval() function for dynamic calculation.
[CREDENTIALS_UNSAFE]: The skill demonstrates how to configure database connections and API keys using appropriate placeholders (e.g., postgresql://user:pass@localhost/langgraph) and environment variable references (e.g., os.environ["LANGCHAIN_API_KEY"]), following standard security practices for credential management.

langchain-architecture