The Agent Skills Directory

SAFE (SAFE): No security issues detected. The skill demonstrates high-quality security hygiene across multiple dimensions.\n
Credential Security: Instructions explicitly forbid sharing API keys in chat and mandate the use of a local .env file or environment variables. This prevents accidental credential leakage into logs or model training data.\n
Network and Download Safety: Downloads are restricted to HTTPS and validated fastfold.ai subdomains. The download script (download_cif.py) also enforces file size limits and prevents HTTP redirects, mitigating potential SSRF or resource exhaustion attacks.\n
Input Validation: The skill uses a dedicated utility module (security_utils.py) to validate UUIDs for job IDs and ensure API base URLs are well-formed and use secure schemes.\n
Mitigation of Indirect Prompt Injection (Category 8): The skill identifies and mitigates the risk of instructions embedded in API data.\n
Ingestion points: API responses from api.fastfold.ai containing job names, sequences, and artifact URLs (fetched in fetch_results.py and wait_for_completion.py).\n
Boundary markers: SKILL.md contains explicit 'Security Guardrails' instructions for the agent to treat all API JSON as untrusted data and ignore embedded instructions.\n
Capability inventory: Scripts perform targeted network requests and file writes but do not use dynamic execution (exec/eval) or shell execution of untrusted strings.\n
Sanitization: Strict validation of IDs and URLs is implemented prior to use in network or file operations.\n
Minimal Attack Surface: All provided scripts rely solely on the Python standard library, eliminating risks from malicious or vulnerable third-party packages.

fold