The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the user to install the @mcp-z/client library from npm. This is a legitimate dependency provided by the skill's author for its primary functionality.
[COMMAND_EXECUTION]: The skill provides examples for spawning local MCP servers using createServerRegistry. This involves executing Node.js scripts as subprocesses to facilitate communication between the client and local server implementations.
[DYNAMIC_EXECUTION]: The library allows for the runtime execution of specified commands and arguments to start MCP servers. Examples demonstrate using process.execPath to launch local script files.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external MCP servers via tool results, prompts, and resources. While this creates a theoretical surface for indirect prompt injection, the risk is inherent to the protocol's purpose and is mitigated by the library's structured response handling.
Ingestion points: External data enters through client.callTool, client.getPrompt, and client.readResource as described in SKILL.md and related examples.
Boundary markers: Explicit boundary markers for LLM prompts are not featured in the provided code examples, though the library uses structured types for data handling.
Capability inventory: The skill facilitates subprocess creation via createServerRegistry (SKILL.md), network communication with HTTP servers (connect-public-server.md), and local file access for configuration (embedded-cluster.md).
Sanitization: The library provides response wrappers that parse content (e.g., .json(), .text()) and enforce typed error handling for incompatible data.

mcp-z-client