メニュー
Use when creating or revising any Claude Code skill, or when the user says "write me a skill for X". Encodes the house conventions: frontmatter style, body structure, when to bundle scripts, and project vs user scope.
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 a question has a factual or best-practice answer but a single response might hallucinate, or when the user asks to "get consensus" or "sanity check this with multiple agents". Spawns N subagents on the same question with varied personas and tallies answer frequency.
Use when a question has real trade-offs and no single right answer: architecture choices, library selection, design direction. Runs a multi-round debate where agents see each other's positions across rounds, converge, and a synthesis is produced. For factual questions, use consensus instead.
Use when triaging an inbox: fetching unprocessed emails, classifying each into a label taxonomy, and applying labels via the provider's API. Provider-agnostic skeleton; scripts/label.py must be wired to your email provider before first use.
Use when a workflow built on an MCP server has stabilized and should become a production skill, or when the user says "convert this MCP workflow to a skill". Produces a skill with a bundled script that calls the underlying API directly, eliminating the per-session MCP token overhead.
Use when a PRD or spec exists and needs to become an executable backlog. Breaks the document into GitHub issues structured as vertical-slice tracer bullets, marks each as HITL or AFK, and files a final QA issue. Requires the gh CLI.
Use when implementing any task that should end in a commit. Encodes the definition of done: understand, implement with red-green-refactor for backend logic, run typecheck and tests until clean, then commit. Also the unit of work for autonomous loops.
| name | write-a-skill |
| description | Use when creating or revising any Claude Code skill, or when the user says "write me a skill for X". Encodes the house conventions: frontmatter style, body structure, when to bundle scripts, and project vs user scope. |
The skill that writes skills. Follow these conventions and every new skill comes out consistent, cheap to load, and discoverable at the right moments.
.claude/skills/<kebab-name>/SKILL.md. YAML frontmatter with exactly name and description. Only the frontmatter loads by default; the body loads on invoke. That asymmetry is the design: a 60-token frontmatter can guard an 8K-token body. Spend effort on the description, keep the body as short as the job allows.
Write it as "Use when..." conditions: the situations, user phrases, and task shapes that should activate the skill. A description that describes ("helps with database work") never fires; one that triggers ("Use when writing or reviewing SQL migrations, or when the user mentions schema changes") fires exactly when needed. Include 2-3 distinct trigger conditions. If the skill is only ever user-invoked by name, the description still helps the model confirm it picked the right one.
Under 50 lines for most skills. If the body needs long reference material (API docs, schemas, examples), move it to a REFERENCE.md in the skill folder and point to it; it loads only when followed.
Deterministic work goes in scripts/ inside the skill folder; judgment stays in prose. Signals you need a script: the same API calls every run, pagination/retries/auth, output that must be exactly formatted. Script contract: CLI flags in, JSON to stdout, errors to stderr with nonzero exit, secrets from env vars by name. If a step needs model judgment mid-script, split the script there: fetch, decide, apply.
.claude/skills/): the skill encodes THIS repo's workflow, references its paths, or teammates should get it via git.~/.claude/skills/): personal, cross-repo, or contains your private conventions.cp.Test in a fresh session: does it trigger on a natural phrasing of the task? Does the procedure run without the author in the room? Iterate on failures; treat reliability like a metric, not a vibe.