The Agent Skills Directory

[SAFE]: The skill is entirely composed of documentation and instructional markdown. It does not include any scripts, executable binaries, or tool definitions that could perform unauthorized actions.
[SAFE]: The content focuses on security-enhancing design patterns, such as defining 'Locked' surfaces for evaluation metrics and 'Human-controlled' surfaces for destructive operations like merging code or deploying to production.
[SAFE]: No obfuscation, data exfiltration patterns, or prompt injection vectors were detected. All external references point to well-known academic or open-source research projects in the AI safety and autonomy space.

harness-engineering