メニュー
Use when reporting task completion, verification results, blockers, failed checks, or unverified work. Require current-task evidence and clearly separate completed, verified, unverified, blocked, and recorded issues.
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。
SOC 職業分類に基づく
| name | proofline-completion-evidence |
| description | Use when reporting task completion, verification results, blockers, failed checks, or unverified work. Require current-task evidence and clearly separate completed, verified, unverified, blocked, and recorded issues. |
Use assets/templates/final-report.md for a detailed completion report and assets/templates/blocker-report.md when work is blocked.
Use current-task evidence before reporting completion:
Separate completed, verified, not verified, blocked, and issues recorded.
If verification cannot run, say why.
Use one concrete line per check.
Never use intention, memory, past success, "looks right", indirect confidence, or unrun checks as completion proof.
Use when a real side issue is found but will not be fixed within the current approved scope. Record only evidence-backed issues that can affect future work and have a concrete suggested next step.
Use when the user asks to port, migrate, copy, or transplant an implementation while preserving behavior exactly or forbidding a rewrite. Keep the source authoritative and require approval for every deviation.
Use for refactors, restructuring, responsibility splits, dependency cleanup, architecture or module-boundary changes, and state or data flow changes. Prove that the intended structure changed rather than only names or surface code.
Use when work is large, risky, multi-step, or easy to shrink accidentally. Keep the full approved goal, checkpoints, verification plan, risks, and approval decisions visible. Not for ordinary low-risk work.
Use for all user-facing responses and artifacts. Write naturally in the user's language, localize ordinary source-language prose, prefer plain wording, clarify consequential ambiguity, make outputs stand alone, apply corrections beyond literal example details, keep diagnostic or rejected wording out of outputs, and keep UI text and code readable.
Use for Codex capability growth: review repeated manual work, shortlist automation candidates, check existing tools, and prepare skill/script/hook/Codex automation registration only after user approval.
