メニュー
Use when the task is multi-constraint, ambiguous, architectural, or needs a design/algorithm choice before answering. Skip for trivial lookups and one-line edits.
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
Use when the user asks to commit, write a commit message, generate a conventional commit, or invokes /conventional-commit. Also use when staged changes need a commitlint-compliant message, when the user asks to commit and push (or passes --push), when unstaged or untracked files exist and staging scope must be confirmed, or when agent-attribution trailers must be avoided on commits. Handles git commit and git push flags when the user passes them.
Use when the user asks to create, open, or raise a GitHub pull request, submit changes for review. Also when staged, committed, or unpushed work needs a PR via gh CLI and required PR metadata is missing (base branch, draft vs ready, reviewers, labels, linked issues, test plan).
Use when the user explicitly asks to audit, critic, or verify a generated solution or answer (e.g. /verifying-solutions, "audit this", "critic this") — not for routine completion claims or pre-answer design.
Use when staging, formatting, or creating git commits for code changes in a Git repository.
Analyzes error logs, stack traces, and provides systematic debugging strategies and solutions
Drafts professional emails based on context, tone, and purpose with templates and suggestions
| name | structured-reasoning |
| description | Use when the task is multi-constraint, ambiguous, architectural, or needs a design/algorithm choice before answering. Skip for trivial lookups and one-line edits. |
Reason in four steps before answering. Precedence: user constraints > correctness > clarity > optimization.
Does not replace diagnose (bugs) or karpathy-guidelines (coding discipline).
Use: multi-constraint design, ambiguous requirements, architecture/algorithm choice, /structured-reasoning / "use structured reasoning", or show work / think out loud.
Skip: trivial lookups, renames, one-line edits, or "be thorough" on a tiny change.
Time pressure does not excuse skipping a step. Recommendation-first then backfill is not this recipe.
| Condition | Output |
|---|---|
| Show work / reasoning / think out loud | ### Thought Process (all 4 labeled steps) then ### Solution |
| Explicit invoke (/structured-reasoning or clear ask) | Same visible template (invoke = show structure) |
| Auto on complex task | Run steps internally; concise answer only — no Task / Thought Process / Solution headings |
"Keep it short" after invoke shortens Solution, not the template.
### Task
[Problem]
### Thought Process
1. Analyze: ...
2. Deconstruct: ...
3. Hypothesize: ...
4. Draft: ...
### Solution
[Answer matching the thought process]
| Excuse | Reality |
|---|---|
| "Just pick something — meeting started" | Run all 4 steps; shorten Solution |
| "Thinking out loud" without labels | Show-work / invoke needs the template |
| "Be thorough" on a rename | Skip skill; ship the one-liner |
| "Keep it short" after invoke | Short Solution; still emit template |
| Recommendation then rationale | Recipe first, then Solution |