The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documentation (references/cli-commands.md) and testing scripts (scripts/test-server.sh) contain process management commands like kill and killall. These are used by developers to clear occupied network ports (e.g., lsof -ti:8000 | xargs kill -9) or stop local test server instances during development.
[EXTERNAL_DOWNLOADS]: The skill references official Python and Node.js packages in its templates and project scaffolding agent, including fastmcp, httpx, and zod. These dependencies are sourced from legitimate public registries (PyPI and NPM).
[PERSISTENCE]: The fastmcp install command documented in references/cli-commands.md facilitates the registration of MCP servers by modifying the local claude_desktop_config.json file. This is a standard and expected mechanism for integrating MCP servers with the Claude desktop client.
[DYNAMIC_EXECUTION]: The utility script scripts/test-server.sh dynamically generates a temporary Python script and executes it to perform automated testing of the MCP server tools and resources. This implementation is local and serves a clear diagnostic purpose.
[INDIRECT_PROMPT_INJECTION]: The templates provided for API and database integration (e.g., api-client-pattern.py, openapi-integration.py) create a surface for indirect prompt injection by ingesting data from external sources.
Ingestion points: External data enters through tools designed to fetch from URLs or endpoints.
Boundary markers: The templates do not implement specific delimiters or warnings to prevent the LLM from executing instructions that might be embedded in the external data.
Capability inventory: The framework supports network requests and file system operations.
Sanitization: The templates encourage the use of Pydantic for strict schema validation of ingested data.

fastmcp