CtrlK
BlogDocsLog inGet started
Tessl Logo

docs-writing-style

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).

58

Quality

67%

Does it follow best practices?

Impact

No eval scenarios have been run

SecuritybySnyk

Passed

No known issues

Optimize this skill with Tessl

npx tessl skill review --optimize ./skills/documentation/docs-writing-style/SKILL.md
SKILL.md
Quality
Evals
Security

Docs writing style

Two modes. Choose one per session.

ModeWhen to useWhat you get
HandoffBefore writing — you want your coding agent (Claude/Cursor/Copilot) to match the existing voiceA self-contained prompt block to paste into your coding agent's system instructions
ReviewAfter drafting — you want to check your draft's tone, structure, and style before submittingA rubric-based diagnosis: what passes, what needs fixing, in priority order

State your mode at the start: "handoff" or "review."


Core principle

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.


Handoff mode

Load references/style-principles.md for the full principles list. Load references/scalekit-style-prompt-block.md (or <docs-repo>/.devex-kit/style-prompt-block.md if the consumer has one) for the site-specific prompt block.

In handoff mode:

  1. Load the style prompt block for this docs site.
  2. Present it directly — the contributor pastes it into their coding agent's system instructions or project-level CLAUDE.md / .cursorrules / .github/copilot-instructions.md.
  3. Optionally: if the contributor describes what they are writing (e.g., "a how-to guide for agent auth"), append a 2–3 line supplement tailoring the prompt to that content type.

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.


Review mode

Load references/style-principles.md for the full principles list. Load references/llms-judge-rubric.md for the quality rubric.

In review mode:

  1. Ask for the draft (a file path, a pasted block, or both).
  2. Run the rubric against the draft.
  3. Report findings in this format:
PASS   <criterion> — <brief note if anything to mention>
FAIL   <criterion> — <specific problem> → fix: <what to do>
WARN   <criterion> — <borderline issue worth watching>
  1. Group by priority: fails first, warns second, passes last.
  2. End with one sentence: the most impactful single fix, if any.

Do not flag style issues in code blocks. Code is reviewed separately.


Key rules (quick reference)

These apply regardless of mode. Violations in review mode are always FAIL.

Structure

  • Sentence case for all headings. No "Configure Your Auth Provider" — use "Configure your auth provider."
  • Page opens with a clear one-paragraph statement: what the reader will do and why.
  • Procedure steps use imperative verbs: "Install", "Run", "Configure" — not "Installing", "You should run."
  • H2 for major sections; H3 for subsections. Never use H1 in body content.
  • Step headings inside <Steps> must be H3 (###) or smaller — never H2. H2 inside Steps creates oversized step titles that break visual hierarchy.
  • Single-item lists are not lists. If a section (e.g. Prerequisites) has only one bullet or numbered item, remove the list markup and write it as a plain paragraph or a smaller heading with prose.

Language

  • Active voice: "Run the command" not "The command should be run."
  • Second person: "you" for instructions, not "the developer" or "users."
  • Present tense: "This method returns" not "This method will return."
  • No filler: no "just", "simply", "basically", "obviously", "quickly", "we're excited."
  • Expand abbreviations on first use: "single sign-on (SSO)" not just "SSO."

Code

  • All four SDK languages (Node.js, Python, Go, Java) in <Tabs syncKey="tech-stack"> for 90% of SDK examples.
  • SDK variable names are NON-NEGOTIABLE: scalekit (Node.js), scalekit_client (Python), scalekitClient (Go/Java).
  • No hardcoded secrets. Environment variables only. Add a comment explaining the security reason.
  • Show error paths alongside the happy path.
  • Examples must be runnable or explicitly marked as pseudocode.

Asides and callouts

  • <Aside> is for incidental context only — cautions, tips, notes.
  • Never put required steps or critical information inside an <Aside>.
  • Use type="caution", type="tip", or type="note" with a title attribute.

Links

  • Descriptive link text only. Never "click here" or "this link."
  • Backticks for code identifiers, file paths, environment variables, and API endpoints.

Did this help?

At the end of every session, ask: "Did this solve what you were trying to do?"

  • If yes: done.
  • If the style prompt was wrong for your site, the review missed something, or the rubric produced unhelpful output: encourage the user to file an issue at https://github.com/saif-shines/devex-kit/issues. Offer to help draft it using their agent — include: which mode they used, what they were reviewing or writing, and what was wrong or missing.

Adapt to your docs site

This skill is site-agnostic. The Scalekit-specific prompt block is in references/scalekit-style-prompt-block.md. To adapt:

  1. Copy references/_template-style-prompt-block.md<your-docs-repo>/.devex-kit/style-prompt-block.md.
  2. Fill in your site's voice, SDK naming conventions, and code standards.
  3. The skill checks for <docs-repo>/.devex-kit/style-prompt-block.md first; falls back to the Scalekit sample with a note that it is an example.
Repository
saif-shines/devex-kit
Last updated
Created

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.