The Agent Skills Directory

[SAFE]: The skill includes a bundled Python script (scripts/official_evaluate.py) for static analysis of skill directories. This script is dependency-free and scans for security patterns like hardcoded credentials without executing untrusted code.
[SAFE]: The instruction set mandates a security review process that looks for malicious patterns, such as exfiltration or unauthorized downloads, in other skills. It includes a specific safety requirement to never reproduce discovered secrets in the evaluation report.
[SAFE]: The skill provides a rigorous evaluation framework based on an 'Official Rubric', which helps prevent the agent from being manipulated by indirect prompt injections contained within the skills being audited.
[SAFE]: All provided external links point to official documentation for the Agent Skills specification and related platform guidance, posing no security risk.

skill-official-evaluation