From documentation
Two-mode writing guide — exports a paste-ready style prompt for coding agents (handoff mode) or reviews a draft against the documentation quality rubric (review mode).
How this skill is triggered — by the user, by Claude, or both
Slash command
/documentation:docs-writing-styleThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Two modes. Choose one per session.
Two modes. Choose one per session.
| Mode | When to use | What you get |
|---|---|---|
| Handoff | Before writing — you want your coding agent (Claude/Cursor/Copilot) to match the existing voice | A self-contained prompt block to paste into your coding agent's system instructions |
| Review | After drafting — you want to check your draft's tone, structure, and style before submitting | A rubric-based diagnosis: what passes, what needs fixing, in priority order |
State your mode at the start: "handoff" or "review."
Documentation voice is a contract with the reader. Inconsistent tone makes docs feel untrustworthy. Consistent tone — even across many contributors — makes readers confident the information is accurate.
Load
references/style-principles.mdfor the full principles list. Loadreferences/scalekit-style-prompt-block.md(or<docs-repo>/.devex-kit/style-prompt-block.mdif the consumer has one) for the site-specific prompt block.
In handoff mode:
.cursorrules / .github/copilot-instructions.md.The prompt block is self-contained. The contributor does not need to explain the docs site to their coding agent — the prompt does it for them.
Load
references/style-principles.mdfor the full principles list. Loadreferences/llms-judge-rubric.mdfor the quality rubric.
In review mode:
PASS <criterion> — <brief note if anything to mention>
FAIL <criterion> — <specific problem> → fix: <what to do>
WARN <criterion> — <borderline issue worth watching>
Do not flag style issues in code blocks. Code is reviewed separately.
These apply regardless of mode. Violations in review mode are always FAIL.
Structure
<Steps> must be H3 (###) or smaller — never H2. H2 inside Steps creates oversized step titles that break visual hierarchy.Language
Code
<Tabs syncKey="tech-stack"> for 90% of SDK examples.scalekit (Node.js), scalekit_client (Python), scalekitClient (Go/Java).Asides and callouts
<Aside> is for incidental context only — cautions, tips, notes.<Aside>.type="caution", type="tip", or type="note" with a title attribute.Links
At the end of every session, ask: "Did this solve what you were trying to do?"
This skill is site-agnostic. The Scalekit-specific prompt block is in references/scalekit-style-prompt-block.md. To adapt:
references/_template-style-prompt-block.md → <your-docs-repo>/.devex-kit/style-prompt-block.md.<docs-repo>/.devex-kit/style-prompt-block.md first; falls back to the Scalekit sample with a note that it is an example.npx claudepluginhub saif-shines/devex-kit --plugin documentationProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.