The Agent Skills Directory

[SAFE]: The skill operates entirely within the user's local project context, specifically targeting the .claude/ directory for any proposed extensions. It explicitly prohibits modifications to the core plugin directory (aura-frog/).
[SAFE]: A mandatory human-in-the-loop confirmation protocol is enforced. The skill is prohibited from creating or writing any files autonomously; it must surface a proposal and wait for explicit user consent before delegating the action to a separate command tool.
[SAFE]: The skill uses a limited set of tools (Read, Glob, Grep) primarily for gathering project context (e.g., reading CLAUDE.md or README files) to ensure proposed extensions are relevant. No network operations or external data exfiltration capabilities were detected.
[SAFE]: Implementation includes a 'detection budget' to prevent user fatigue, limiting proposals to one per turn and three per session.
[SAFE]: The skill does not contain any obfuscated code, hidden instructions, or attempts to bypass agent safety guidelines.

extension-detector