The Agent Skills Directory

[SAFE]: The skill establishes a bounded development lifecycle with explicit iteration budgets and stop conditions, preventing uncontrolled or runaway agent behavior.
[SAFE]: Instructions mandate human approval gates for sensitive operations, including destructive commands, production deployments, and external service writes, maintaining human-in-the-loop control for high-risk tasks.
[SAFE]: Security-centric guidelines are integrated into the workflow, specifically advising the agent to avoid hardcoded credentials and to execute generated tests within sandboxed environments.
[SAFE]: The implementation phase is strictly constrained to the approved specification, preventing the agent from performing unauthorized refactorings or adding undocumented features outside the user-defined scope.

ai-loop