The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external Model Context Protocol (MCP) servers, which serves as a potential vector for indirect prompt injection.
Ingestion points: Data is pulled into the agent's context through functions such as load_mcp_tools(), load_mcp_resources(), and load_mcp_prompt() described in SKILL.md.
Capability inventory: The skill converts remote MCP tools into local LangChain BaseTool objects, enabling the agent to execute code or operations provided by the server. Resources can also be loaded as Blob objects.
Sanitization: No explicit sanitization or input validation mechanisms for server-provided content are described in the documentation.
Boundary markers: The skill does not provide instructions or examples for implementing delimiters to separate untrusted server data from system instructions.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the installation of the langchain-mcp-adapters package from public registries.
[COMMAND_EXECUTION]: The documentation includes examples of spawning local processes using stdio transport, specifically demonstrating the use of npx to launch MCP servers such as @modelcontextprotocol/server-filesystem.

langchain-mcp-adapters