The Agent Skills Directory

[SAFE]: Audit Orchestration Logic. The skill functions as a thin orchestrator for project audits. Its primary activities consist of reading local documentation (e.g., docs/project-config.json, docs/project-reference/) and writing reports to the project's local plans/reports/ directory. No external network exfiltration or unauthorized file access was observed.
[SAFE]: Sub-agent and Protocol Management. The skill spawns platform-internal sub-agents and uses internal commands (such as $project-init or $why-review) to manage the review workflow. It includes mandatory validation gates to verify evidence and deduplicate findings, promoting accuracy and safety.
[SAFE]: Instruction and Prompt Integrity. The instructions emphasize strict task tracking, evidence-based reasoning, and adherence to established protocols. While it contains instructions to disregard legacy platform-specific formatting notes, it does not attempt to bypass core AI safety filters or exfiltrate sensitive user data.

architecture-review-full