The Agent Skills Directory

[SAFE]: The skill follows a strict 'secretary' persona that prioritizes human-in-the-loop oversight. All proposed changes are marked as 'PROPOSED' and the agent is instructed to wait for explicit confirmation before modifying any project state.
[PROMPT_INJECTION]: The skill processes untrusted user data (weekly review notes and charters), creating a potential surface for indirect prompt injection. However, this risk is mitigated by the 'grant' system and the 'PROPOSED' workflow defined in ROLE.md, which prevents the agent from executing instructions found in that data without the user's manual intervention.
[DATA_EXFILTRATION]: No network operations or unauthorized file access patterns were detected. The skill operates exclusively on the local markdown vault using configuration-resolved paths.

review-quarter