The Agent Skills Directory

[SAFE]: The skill provides purely instructional content without any executable scripts or binary files, minimizing the direct attack surface.
[SAFE]: It implements a 'Review Critically' step before execution, requiring the agent to identify and resolve concerns with a human partner, which mitigates risks associated with untrusted plan inputs.
[SAFE]: The skill establishes clear boundaries for autonomous action by requiring explicit user consent for modifications to primary branches and providing strict guidelines on when to stop execution.

executing-plans