Skill 詳細
make-documentation
Writes and audits broad project documentation deliverables.
使用前に確認
自動レビューは関連性のみを確認し、安全性や推奨を保証しません。使用前に出典の説明を読んでください。
SKILL.md
これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。
--- name: make-documentation description: "Use when writing READMEs, architecture, changelogs, release notes, runbooks, notebook docs, install docs, or security articles; produces docs. Do not trigger for API docs, AGENTS.md, or diagrams." metadata: short-description: Write project documentation. --- # Make Documentation ## When to Use - Writing or updating a README.md, architecture note, changelog, or release notes - Auditing existing project docs before adding missing documentation - Drafting customer-facing security guidance or partner-sensitive article prose from source material - Writing live access, reproduction, lab interface, workstation/server, or operator runbooks - Documenting notebooks, generated notebooks, or table-producing notebook workflows - Writing WSL, local, Docker, or environment-specific install docs - Writing agent operation manuals, role files, peer-status contracts, or AI-readable workflow docs - Producing diagram documents only when the user explicitly asks or the project already uses them ## When NOT to Use - Writing API reference documentation - Creating or updating AGENTS.md. Use `agents-md`, even when the broader request also includes other docs. - Generating diagram files by default when prose or tables are sufficient ## Workflow 1. Run [scripts/audit_documentation.py](scripts/audit_documentation.py) for new docs, large rewrites, doc moves, missing-doc investigations, or unclear structure. For tiny explicit edits, inspect the target file and nearby docs directly. 2. Choose only the deliverables justified by the user's request and the audit. Do not generate a fixed documentation bundle by default. 3. Route by document type: use [references/guide-readme.md](references/guide-readme.md) for `README.md`, [references/guide-architecture.md](references/guide-architecture.md) for `docs/architecture.md` or the repo's existing equivalent, [references/guide-ai-ingestion.md](references/guide-ai-ingestion.md) when the user says the document is for agents/AI ingestion, [references/guide-agent-ops-docs.md](references/guide-agent-ops-docs.md) for role/workflow docs consumed by agents, [references/guide-access-runbook.md](references/guide-access-runbook.md) for live access, reproduction, operator, lab interface, or capability ledgers, [references/guide-notebook-documentation.md](references/guide-notebook-documentation.md) for notebook explanations, [references/guide-install-runbook.md](references/guide-install-runbook.md) for WSL/local/Docker install docs, [references/guide-security-article.md](references/guide-security-article.md) for partner-sensitive security articles, and [references/guide-changelog.md](references/guide-changelog.md) for `CHANGELOG.md` or release notes. If AGENTS.md becomes part of scope, switch that file to `agents-md` instead of expanding this skill. 4. Use [references/guide-diagrams.md](references/guide-diagrams.md) only when the user explicitly asks for diagram docs or the project already maintains them. 5. For large docs, iterative reports, feature inventories, concept framing, or evidence-audit amendments, load [references/source-discovery-and-heading-stability.md](references/source-discovery-and-heading-stability.md) before writing. 6. For concept or strategy docs, include a visible first usable workflow before architecture depth so the reader can execute a small slice before absorbing the model. 7. Keep the output source-backed and terse. Preserve user-provided terminology and avoid generic rewrites that change domain meaning or connotation. 8. Preserve earlier requested documentation when later implementation work touches the same notebooks, docs, or folders. Re-open touched docs before finishing if there is a risk an unrelated implementation change removed prior documentation. 9. Review the result with [references/review-checklist.md](references/review-checklist.md) before finishing. ## Anti-Patterns - Wrong: summarize a notebook from a generated README, review packet, AIGitHub で全文を読む (外部ページ)