The Agent Skills Directory

[SAFE]: The skill performs legitimate file operations restricted to the local project directory (e.g., reading hypotheses and writing PRDs to docs/prd/). No unauthorized file access or exfiltration patterns were detected.
[SAFE]: The execution protocol includes multiple [STOP — BLOCKING] markers that require explicit user approval before proceeding, providing a strong safeguard against unintended automated actions.
[SAFE]: The workflow uses a separate doc-reviewer sub-agent to assess quality, which is a standard pattern for multi-step task separation and does not introduce new security vulnerabilities.
[SAFE]: No obfuscation, remote code execution, or suspicious network activities (such as calls to external domains or credential harvesting) are present in the skill definition.

recipe-define