The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to manage and interact with MCP servers. This involves spawning subprocesses to run server executables as defined in the local configuration file .claude/.mcp.json. This behavior is implemented in scripts/mcp-client.ts using the @modelcontextprotocol/sdk.
[EXTERNAL_DOWNLOADS]: The documentation in README.md and references/configuration.md recommends running various MCP servers (such as @modelcontextprotocol/server-memory and server-filesystem) using npx -y. This command pulls the latest versions of these packages from the npm registry during execution.
[REMOTE_CODE_EXECUTION]: The assets/tools.json metadata file includes definitions for high-capability tools like evaluate_script from the chrome-devtools server and eval equivalents in other servers. These tools allow for the execution of arbitrary code within their respective sandboxed environments (e.g., a browser or an isolated node process), which is an intended capability for these specific developer tools.

mcp-management