The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of several packages via pip, including langchain, langchain-anthropic, langgraph, and langchain-mcp-adapters. These are well-known libraries maintained by the LangChain organization.
[INDIRECT_PROMPT_INJECTION]: The skill documentation (specifically in references/retrieval.md) outlines patterns for building agents that ingest data from external sources such as web pages (WebBaseLoader), PDFs (PyPDFLoader), and vector databases. This creates a surface for indirect prompt injection where malicious content in the retrieved data could influence the agent's behavior.
Ingestion points: references/retrieval.md uses WebBaseLoader, PyPDFLoader, and vectorstore.similarity_search to pull external content into the agent's context.
Boundary markers: The provided code templates do not include explicit delimiters (e.g., XML tags or triple backticks) or "ignore embedded instructions" warnings around the retrieved content.
Capability inventory: The agents described in SKILL.md and the reference solution have capabilities to execute tools like search_orders, submit_refund, and get_account based on LLM decisions.
Sanitization: The example code does not demonstrate sanitization or validation of the retrieved text before it is presented to the model.
[COMMAND_EXECUTION]: In SKILL.md, the documentation for MCP (Model Context Protocol) integration demonstrates how to configure the MultiServerMCPClient to execute local shell commands or scripts (e.g., python /path/to/server.py). This feature allows the agent to interact with local system tools.

langchain