Agent Stylebooks
Pick the writing system before the model picks one for you.
Stop asking AI to “write better.” Give it an editorial system.
Portable Agent Skills based on established editorial systems.
npx skills add Neeeophytee/agent-stylebooks
Browse stylebooks · Installation · Examples · Provenance
Each stylebook defines what to lead with, how to order information, which distinctions must survive editing, what ambiguity is unacceptable, and how to check the result. These are community-maintained Agent Skills derived from publicly documented editorial guidance, not official vendor or agency products.
Which stylebook should I use?
Choose by the reader's task and the artifact—not by the organization named in the draft.
| Recommended stylebook | I am writing... |
|---|---|
$google-developer-docs | API or setup tutorial, how-to guide, onboarding instructions |
$govuk | Public-service eligibility page, decision guide, application instructions |
$gitlab-docs | Engineering or product docs, internal documentation, feature guide |
$github-docs | Product workflow, step-by-step guide, troubleshooting article |
$mdn-web-docs | Web API explanation, technical reference, learning article |
$kubernetes-docs | Infrastructure procedure, operations runbook, deployment guide |
$nhs-health-content | Health or patient content, explainer, care instructions |
$cdc-clear-communication | Public-health message, safety advisory, awareness campaign |
$sec-plain-english | Investor disclosure, business report, risk explanation |
$w3c-technical-reports | Technical specification, standards document, requirements definition |
$nasa-technical-writing | Engineering or test report, research report, findings summary |
$microsoft-writing-style | Product help or UX copy, support article, interface guidance |
$apple-interface-writing | Interface labels or alerts, microcopy, onboarding flow |
Prompt vs stylebook
| Generic prompt | Stylebook |
|---|---|
| Ad hoc | Reusable |
| “Make it clear” | Concrete editorial decisions |
| Source usually unknown | Provenance documented |
| Rewritten each session | Installed once |
| Easy to drift | Repeatable workflow |
| “Sound human” | Optimize for a specific reader and task |
See the difference
These original examples illustrate the systems; they are not quotations or claimed outputs from the source organizations.
Generic technical prose
To successfully initiate configuration, authentication credentials must first be appropriately established.
$google-developer-docs
Before you configure the integration, set up your authentication credentials.
Generic error
An error occurred while processing your request. Please try again later.
$microsoft-writing-style
We couldn't save your changes. Check your connection and try again.
Generic risk statement
Unfavorable market conditions may adversely affect our results.
$sec-plain-english
If borrowing costs rise, our interest expense may increase and reduce our net income.
Install
The portable installer currently requires Node.js 22.20 or newer.
# List the catalog
npx skills add Neeeophytee/agent-stylebooks --list
# Install one stylebook
npx skills add Neeeophytee/agent-stylebooks --skill cdc-clear-communication
# Install all 16
npx skills add Neeeophytee/agent-stylebooks
Claude Code and Codex can also install the repository as a plugin. Manual, project-level, and agent-specific routes are in INSTALL.md.
Use
Use $google-developer-docs to turn these notes into a setup guide.
Use $govuk to rewrite this eligibility page around the user's decision.
Use $apple-interface-writing to tighten these labels and error messages.
Use one primary stylebook unless the task explicitly needs a hybrid. A product terminology guide still governs product names; the stylebook governs structure and prose. Every skill must preserve supplied facts, code behavior, legal meaning, and domain constraints.
Compatibility
Every stylebook uses the open SKILL.md format and remains an immediate child of skills/.
That flat layout supports Codex, Claude Code, Kimi Code CLI, Cursor, Hermes Agent, GitHub
Copilot, Gemini CLI, and other compatible loaders.
| Agent | Supported route |
|---|---|
| OpenAI Codex | Codex plugin or .agents/skills/ |
| Claude Code | Claude plugin or .claude/skills/ |
| Kimi Code CLI | Native Agent Skills directories or portable installer |
| Cursor | Portable installer or native project skills directory |
| Hermes Agent | skills.external_dirs or ~/.hermes/skills/ |
| GitHub Copilot | .github/skills/, .agents/skills/, or .claude/skills/ |
| Gemini CLI | .gemini/skills/ or portable installer |
| Other compatible agents | Any loader that implements the Agent Skills specification |
Compatibility means the repository follows each documented discovery shape. It does not imply that the source organizations created or tested these stylebooks. See COMPATIBILITY.md for test evidence and INSTALL.md for the exact routes.
Provenance and limits
PROVENANCE.md records the authoritative source and reuse decision for every stylebook. Openly licensed and public-domain sources still use independently written rules. Restricted or unclear sources are reference-only interpretations.
The repository's original text and scripts are MIT licensed. Third-party guides retain their own terms. No source organization endorses this project. A stylebook does not provide legal, medical, regulatory, standards, or publication approval. See NOTICE.md.
Validate and contribute
python3 scripts/validate_repo.py
CI checks structure, names, frontmatter, manifests, provenance, links, duplicate inventory, and the byte integrity of the original 11 skills. Read CONTRIBUTING.md before proposing another editorial system or use the stylebook request form.