goberomsu/craft-skills/document
Claude Code and agent skill by goberomsu/craft-skills
- community
Checking that the SKILL.md file is still in the repository.
Scaffolds and authors repository documentation through a six-type ontology and canonical artifacts. Use when asked to scaffold repository docs, "record this decision", "where does this spec go", "draft the project CHANGELOG", or "comment-the-why". Not for READMEs (use readme), conducting research (use research), technical reports (use write-report), API-surface comments, or DESIGN.md, visual direction, UX audits, and design-system work (use design).
Install
npx vskill@latest install goberomsu/craft-skills/documentSKILL.md
document
Author and file project documentation so every artifact lands at exactly one canonical location with the right template. Success: a reader or agent can always answer "where does this go?" without guessing. Run unslop as the final pass over the prose this skill writes; code and fixed machine formats are excluded.
Requirements
- POSIX
find,sed,grep(BSD or GNU) — the root-doc-sprawl check below.
Artifact types (MECE)
Every documentation artifact answers exactly one question; never let one artifact try to answer two.
| Artifact | Question answered | Lifespan | Template | Canonical location | |
|---|---|---|---|---|---|
| research | What did I find — facts, sources, comparisons (pre-decision) | Topic-scoped; superseded as evidence evolves | templates/research.md | docs/research/{slug}.md | |
| references | What does this external source say — verbatim static archive of a third-party document | Permanent snapshot; never edited after capture | templates/references.md | docs/research/references/{slug}.md | |
| spec | What is this work and are the requirements clear? | Work-scoped, one-shot | templates/spec.md | docs/exec-plan/active/{slug}/spec.md | |
| plan | How to implement (steps, files, order) | Work-scoped; body immutable only after explicit finalization | templates/plan.md | docs/exec-plan/active/{slug}/plan.md | |
| decision | One expensive-to-reverse cross-cutting decision | Permanent topic anchor; edited in place, each change logged in the ADR's ## Changelog | templates/adr.md | docs/decisions/ADR-NNN-{topic}.md | |
| rule | A standing convention (ongoing constraint, not work-scoped) | Alive as long as the convention holds | templates/rule.md | docs/rules/{topic}.md |
Routing
Pick the primary question a task answers and route by that row — the ontology table above for the six ontology artifacts, or the table below for repo-level artifacts and standing conventions. Create a decision artifact only when the user explicitly asks to record a decision or write an ADR; otherwise keep research, specs, plans, and rules in their own lanes and note decision candidates inside that artifact.
| Task | Read | Author with | |
|---|---|---|---|
| Record a cross-cutting decision / write an ADR | references/adr.md | templates/adr.md | |
| Write the project-level CHANGELOG / release notes | references/changelog.md | templates/changelog.md | |
| Decide which code comments survive (Comment Sicko keep-list) | references/inline-comments.md | — | |
| Write or review prose in a doc, RFC, README, PR description, or commit message | references/technical-writing.md | — | |
Map the system (architecture.md) | see architecture.md section below | templates/architecture.md | |
| File research, a spec, a plan, a rule, or an archived source | ontology table above | matching file in templates/ | |
| Scaffold repository documentation | docs/ layout below | create the applicable canonical directories and artifacts |
No documentation task should dead-end here — every repo-level artifact and every ontology artifact has an exit row.
Nearest-neighbor boundary
design owns DESIGN.md, visual direction, UX audits, and design-system work; route those requests there. readme owns the repository README.md; route README authoring there.
References vs research: a references file is a verbatim static copy of an external document; a research file is your own synthesis. Never merge them.
Rule vs ADR: the what to do goes in a rule. Load references/adr.md only when the user explicitly asks for an ADR or decision record.
Root-doc-sprawl check
Loose .md files accumulate at the repo root instead of filing into the ontology above. Count them, excluding the conventional set:
More than 3 is a signal to investigate the loose documents and propose their canonical docs/ paths before adding more. Do not move existing documents without an explicit request.
Decision boundaries
Research collects evidence and presents options; it does not decide. A plan sequences implementation; it may mention trade-offs without spawning a decision record. An ADR is an explicit destination, not an automatic distillation step. When routing is unclear, ask which artifact the user wants: evidence, a recorded decision, or implementation sequencing.
docs/ layout
Scaffold only the canonical directories and repository documentation artifacts requested; do not create empty placeholder files or move existing documents without an explicit request.
Scratch drafts live outside docs/ and are not git-canonical. Finalizing a draft means moving it into docs/ at the correct path, then deleting the scratch source.
Lifecycle — spec and plan
Lifecycle = folder position. A plan is active or archived by where its folder lives, never by a status field alone; move the whole {slug}/ folder, never split spec and plan.
Supersede timing: when creating a superseding plan-B, archive plan-A in the same action, before beginning plan-B's execution.
Finalized-plan contract: a working plan remains editable after commits. Set finalized: true only when the user explicitly adopts its body as immutable; from then, a scope change gets a new slug and the old plan's frontmatter is updated to status: superseded + superseded-by: {new-slug}. Only the finalization and lifecycle frontmatter changes after finalization.
Slug naming
Format: {kebab-description} — lowercase, hyphens only, no date prefix, no issue number. The folder (or filename) is the authoritative slug; a frontmatter slug that disagrees with it is an error.
architecture.md
Answers "how is this system put together?" for someone who just opened the repo: an annotated directory tree, component boundaries, a cross-cutting-decisions index (links to ADRs, not restatements), and the one or two data/control flows that matter most.
Update it whenever a change moves a component boundary or a new cross-cutting decision lands — not on every commit. Staleness signal: its directory tree or ADR index no longer matches reality. Keep it a map: if a section starts duplicating an ADR body, replace the duplication with a link.
Trivial exemptions — no spec/plan required
- Typo, comment, or wording fixes
- Lint- or format-only changes
- Dependency patch-version bumps (
x.y.Z) - Behavior-preserving renames
Anti-patterns
- Implicitly creating or scheduling an ADR because research, a plan, or a durable choice mentions a decision → keep the original artifact in its lane and create an ADR only when the user explicitly asks to record one.
Verification
- [ ] Each artifact answers exactly one question (an ontology row or a routing-table row) — none dead-ended
- [ ] Any ADR was created only from an explicit ADR / decision-record request
- [ ] Completed/discarded work moved
active/→archive/with astatusline - [ ]
architecture.mdreflects the latest component boundaries and decisions, or was flagged stale - [ ] References files hold verbatim source content, not synthesis
- [ ] Slug matches its folder/filename exactly
- [ ] Root-doc-sprawl check run when loose root docs are suspected; any relocation was proposed and explicitly requested before moving files
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/document/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/goberomsu/craft-skills/document)