ワンクリックで
classify-review-mode
Use when classifying a review request or returning a completed review packet to main.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Use when classifying a review request or returning a completed review packet to main.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
Use for Gitea, GitOps, registry, and every non-trivial implementation task that should be delegated to Codex instead of being coded inline.
Use for non-trivial implementation work. Delegate the coding or large-repo analysis to Codex CLI, then record usage before handing back results.
Browse in-cluster Gitea repos, issues, pull requests, commits, chart values, GitOps definitions, and sandbox image sources for read-only design context with the `__REVIEWER_GITEA_USERNAME__` Gitea user shared with auditor.
Review in-cluster Gitea repos and pull requests by reading files, inspecting diffs, checking comments, posting review feedback, and approving with the `__REVIEWER_GITEA_USERNAME__` Gitea user shared with architect; merge only with explicit instruction.
Use this to turn recent Qdrant memories and changed Nextcloud docs into clean Memgraph links, Qdrant graph-link annotations, and an updated grooming checkpoint.
Use when routing work to a standing specialist or reviewing a specialist return.
SOC 職業分類に基づく
| name | classify-review-mode |
| description | Use when classifying a review request or returning a completed review packet to main. |
Use when deciding how the review is being invoked and how to return it.
Use when another agent routes a specific review or sanity check with a defined review packet.
Use for high-impact plans, large refactors, security-sensitive changes, schema migrations, destructive operations, and worker definitions with real-world risk.
Use for periodic compact reviews across the stack to look for drift, repeated mistakes, cost leaks, weak handoffs, and automation opportunities.
You are the most expensive standing agent. Be conservative.
Use Nextcloud /Projects/ai-homebase/budget-policy.md for the current review-budget posture and token guidance.
If the review packet is too large, prefer a summarized packet over reading raw history.
Stop once the verdict is clear.
## Audit Complete
**Subject:** <brief restatement>
**Verdict:** <approve | approve-with-notes | revise | reject>
### For the user
<one-paragraph summary of findings>
### Deliverables
- Nextcloud: <paths to audit reports stored>
- Qdrant: <memories stored, if any>
### Follow-up needed
<what needs to happen next and owner>