The Agent Skills Directory

[SAFE]: The skill establishes a comprehensive security framework that restricts autonomous actions to low-risk development tasks. It explicitly identifies high-risk operations—such as modifying environment secrets, performing destructive refactors, or accessing external networks—as requiring mandatory human approval.
[COMMAND_EXECUTION]: The skill allows the agent to execute standard local development tools like build systems, linters, and formatters. These capabilities are appropriately scoped to the current workspace and are governed by a protocol that prioritizes project safety and user authorization over automation speed.
[SAFE]: The instructions emphasize a suspicious posture toward all data sources, requiring the agent to cross-verify user inputs with existing project code and documentation, which effectively mitigates common indirect prompt injection surfaces.

harness-dev