The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of external MCP servers as subprocesses. It utilizes the @modelcontextprotocol/sdk to manage connections and tool calls via StdioClientTransport. The specific commands and arguments executed are sourced from a local configuration file (.claude/.mcp.json) which is managed by the user.
[EXTERNAL_DOWNLOADS]: The skill instructions guide users to install Node.js dependencies using npm install and recommend the installation of the gemini-cli tool for enhanced integration. These are documented steps for utilizing standard MCP tooling.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its core function of ingesting metadata from external servers.
Ingestion points: The skill dynamically fetches tool, prompt, and resource definitions from configured MCP servers in scripts/mcp-client.ts.
Boundary markers: No specific boundary markers are identified in the prompt templates that would delimit server-provided tool descriptions from the agent's internal instructions.
Capability inventory: The skill enables the execution of any capability exposed by the configured servers, which can include powerful operations like file system manipulation or browser automation.
Sanitization: Descriptions and schemas received from servers are stored and presented to the LLM without sanitization, relying on the trust established with the configured server endpoints.

mcp-management