The Agent Skills Directory

[SAFE]: The skill implements a diagnostic command (/orient-gaps) to analyze project state without modifying any files or environment settings.
[SAFE]: It utilizes a "bounded read-only subagent" for data gathering, which limits the scope of investigation and prevents unauthorized changes to the system during the discovery phase.
[SAFE]: The skill includes explicit security and reliability rules, such as rule:fail-closed, which ensures the agent reports uncertainty rather than fabricating status when data is insufficient.
[SAFE]: Tool usage for visualization (e.g., mcp__visualize__show_widget) follows platform-standard practices and includes constraints against potentially problematic CSS (e.g., no fixed positioning or hardcoded colors).
[SAFE]: The "self-improvement" capability is designed with a human-in-the-loop requirement (ask-gate), ensuring that any proposed modifications to the skill must be explicitly approved by the user.

orient-gaps