The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch live documentation from https://gofastmcp.com/llms-full.txt and use WebSearch to supplement bundled references. This is documented neutrally as it targets the official domain for the framework being implemented.
[REMOTE_CODE_EXECUTION]: The skill provides templates and guidance for using create_proxy and FileSystemProvider. These features enable the execution of local scripts via subprocesses or the dynamic loading of Python modules from a directory. While these are core features of the FastMCP framework, they represent a capability for local code execution if configured to point at malicious or untrusted files.
[PROMPT_INJECTION]: The skill includes 'Critical Rules' and imperative instructions to guide the agent's behavior. While these are intended for quality control (e.g., 'No stdout', 'Verbose descriptions'), they are structurally similar to override patterns. Additionally, the ingestion of external documentation via WebFetch creates a surface for indirect prompt injection, where instructions embedded in the remote text could influence the agent's implementation strategy.
Ingestion points: The agent fetches implementation details from https://gofastmcp.com/llms-full.txt (SKILL.md § Consult Live Documentation).
Boundary markers: None explicitly defined for the remote content; the agent is simply instructed to let live docs 'win' over bundled references.
Capability inventory: The agent can scaffold new projects using wagents new mcp and write implementation code to the local filesystem.
Sanitization: No explicit sanitization or filtering is performed on the fetched documentation before processing.

mcp-creator