The Agent Skills Directory

[SAFE]: The skill demonstrates high security maturity by using only the Python standard library, avoiding external dependencies, and restricting network communication to the official NCATS ARAX production endpoint. It includes a custom redirect handler to prevent cross-origin and protocol-downgrade attacks.
[INDIRECT_PROMPT_INJECTION]: The skill processes external biomedical data from the ARAX API, which represents an indirect prompt injection attack surface.
Ingestion points: Data is fetched via HTTP GET/POST from the arax.transltr.io domain in scripts/arax_client.py and can also be loaded from local files via the summarize command.
Boundary markers: The client enforces strict TRAPI schema compliance and uses a robust JSON decoder that rejects non-finite values to prevent exploitation of the underlying parser.
Capability inventory: The skill is restricted to fetching scientific data and writing artifacts to a dedicated output directory. It contains no capabilities for arbitrary command execution, dynamic code evaluation, or sensitive system file access.
Sanitization: All text retrieved from external sources is passed through a _sanitize_text function that removes control characters and enforces character length limits before display.
[COMMAND_EXECUTION]: The skill executes its own internal Python script via Bash. The script uses secure coding practices, including strict argument parsing via argparse and atomic file operations for output artifacts.

ncats-arax