The Agent Skills Directory

[COMMAND_EXECUTION]: The mcp-stdio.ts script is designed to spawn child processes to run MCP servers using commands provided by the agent or user. This is the primary intended functionality of the skill.
[EXTERNAL_DOWNLOADS]: The skill facilitates downloading and executing official MCP server implementations and the @modelcontextprotocol/sdk via standard package managers like npm and npx. These are well-known resources within the MCP ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes tool schemas and execution results from external MCP servers, creating a potential surface for indirect injection if a server provides malicious content.
Ingestion points: Data is ingested through HTTP response bodies in mcp-http.ts and standard output streams in mcp-stdio.ts from external servers.
Boundary markers: The skill does not implement explicit delimiters or warnings to the agent to ignore instructions embedded within the tool data received from servers.
Capability inventory: The skill possesses the ability to execute shell commands, perform network requests, and generate new skill files.
Sanitization: The scripts pass the raw JSON data received from the external servers directly to the agent without filtering or sanitization of string content.

converting-mcps-to-skills