The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill makes outbound HTTPS requests to the official NCATS ARAX production API (arax.transltr.io) to retrieve biomedical knowledge-graph data. This is a well-known scientific service and the core purpose of the skill. The implementation includes strict URL validation, rejecting localhost, private IP ranges, and protocol downgrades.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute its own internal Python script (arax_client.py). The script itself does not spawn shell subprocesses or execute user-provided strings as commands; it is a purely instructional wrapper for the API client.
[DATA_EXFILTRATION]: While the skill sends data to a remote API, it implements extensive safeguards: it requires explicit user acknowledgment (--acknowledge-public-query), warns that metadata is public, and strictly forbids the submission of sensitive, proprietary, or patient data. The network operations are limited to the specific biomedical lookup tasks described.
[PROMPT_INJECTION]: The SKILL.md contains robust safety instructions and negative constraints (e.g., 'Do not use for clinical guidance', 'Report zero as not returned'). These are defensive measures to prevent the agent from misinterpreting scientific data, rather than malicious injection patterns.
[CREDENTIALS_UNSAFE]: The skill does not use API keys or hardcoded secrets. It specifically notes in its documentation that the client uses only the standard library and needs no credentials for the public API.
[SAFE]: The skill follows secure coding best practices, including atomic file writes with os.replace, restricted file permissions (0o600), and strict JSON serialization to prevent injection or corruption of local artifacts.

ncats-arax