The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides examples of spawning local MCP server processes as part of its core functionality for managing stdio-based servers.
Evidence: resources/automated-testing.md demonstrates spawning a server using process.execPath with bin/server.js.
Evidence: resources/spawn-local-server.md demonstrates spawning a server using process.execPath with my-mcp-server.js.
[EXTERNAL_DOWNLOADS]: The documentation references a public MCP server endpoint for a well-known service for demonstration purposes.
Evidence: resources/connect-public-server.md and resources/discover-capabilities.md reference https://ai.todoist.net/mcp.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external MCP servers into the agent's context, which is the primary function of an MCP client library.
Ingestion points: client.callTool(), client.getPrompt(), and client.readResource() in SKILL.md allow the agent to receive data from external servers.
Boundary markers: The documentation in SKILL.md mentions response wrappers that handle MCP results and throw typed errors for incompatible content.
Capability inventory: The skill can execute local commands (via registry config), make network calls (to HTTP servers), and access configuration files (like .mcp.json).
Sanitization: The skill describes the use of convenience methods and response wrappers to handle server outputs.

mcp-z-client