GitHub Skill Curator reviews other Agent Skills, so safety matters.

The main branch is the active version.

Please open a GitHub issue if you find:

• secret or credential leakage;
• unsafe install behavior;
• destructive commands without clear user intent;
• prompt-injection patterns that try to override higher-priority instructions;
• scripts that hide behavior through obfuscation, encoding, or minification;
• broad skill descriptions that could hijack unrelated tasks.

Do not include real secrets, tokens, private files, browser profiles, or personal data in the report.

This project provides heuristic scans and governance rules. It cannot prove a third-party skill is safe. It helps users review candidates before installation and keep their local skill folder auditable.