The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses fastmcp CLI and Python subprocesses (via scaffold scripts) to install and run MCP servers. While these are documented developer workflows, they involve executing user-specified or scaffolded Python code on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect injection surface through its templates. The api_wrapper.py and file_processor.py templates ingest untrusted data from external REST APIs and local text files respectively.
Ingestion points: _request function in api_wrapper.py and _read_text function in file_processor.py.
Capability inventory: The templates support network requests (httpx), local file system reads (pathlib), and database queries (sqlite3).
Sanitization: There is minimal sanitization of the content fetched from APIs or files before it is returned to the agent context, which could allow malicious instructions embedded in those sources to influence the agent.
Boundary markers: None identified in the provided templates to separate untrusted data from instructions.
[EXTERNAL_DOWNLOADS]: The skill instructions and prerequisites require installing external packages (fastmcp, httpx) via pip. These are well-known packages, but the skill facilitates their installation and execution.
[DATA_EXPOSURE]: The database_server.py template facilitates querying local SQLite databases. While it attempts to enforce a SELECT only policy, it exposes the full schema and contents of any database provided via environment variables.

fastmcp