product-content-strategist

A Claude Code / agent skill that turns your assistant into a senior product content strategist. It writes and audits in-product copy — UI labels, buttons, microcopy, error messages, empty states, onboarding, settings, notifications, AI feature copy, and feature naming — following established product/UX writing principles from real US design systems.

Not for marketing copy (blogs, ads, landing pages, email campaigns). This is product content only.

Install

Option 1 — Universal (55+ agents)

npx skills add xzjh/product-content-strategist

Or with explicit scope/agent:

npx skills add xzjh/product-content-strategist -g -a claude-code

Works with 55+ agents including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Kiro CLI, and more.

Option 2 — Claude Code plugin marketplace

This repo is also a Claude Code plugin marketplace. Inside Claude Code:

/plugin marketplace add xzjh/product-content-strategist
/plugin install product-content-strategist@xzjh-skills

Updates are pulled with:

/plugin marketplace update xzjh-skills

Option 3 — Manual

Download SKILL.md and drop it into ~/.claude/skills/product-content-strategist/SKILL.md (or your agent's equivalent skills directory).

What it does

The skill encodes the cross-source consensus from authoritative US-based design system documentation, then applies it. Specifically:

• 13 operating principles (plain language, sentence case, active voice, "you", contractions, "select" not "click", inclusive by default, content before wireframes, accessibility, localization-ready, etc.)
• Word blocklist that flags problem language on every audit (errors: invalid, please, sorry, oops, forgot; ableist: crazy, sanity check, easy, simple; tech/race-coded: blacklist, master, grandfathered; gendered, militaristic, cultural appropriation)
• Pattern library for 9 surfaces — error messages, empty states, buttons & CTAs, form labels, notifications & toasts, onboarding, settings & toggles, naming, numbers/dates — each with verbatim sourced ✅/❌ examples and URLs
• AI feature content section — labeling AI presence, set-expectation copy, accept/edit/regenerate/dismiss patterns, anthropomorphization ban, "why this?" rationales, failure fallbacks, banner-blindness caveats
• Voice & tone framework with a Podmajersky-style voice chart the skill produces when no brand voice is documented
• Workflows for both writing tasks (clarify surface → user → action → 2-3 options + tradeoffs) and audits (scan against the blocklist + accessibility + WCAG 2.2 implications)

Sources

All examples in the skill are verbatim quotes from public US design system documentation, with deep-link URLs:

• Shopify Polaris — error messages, voice & tone, inclusive language, actionable language
• Microsoft Writing Style Guide — top-10 tips, bias-free communication, A-Z word list
• Microsoft Windows Admin Center UI Style Guide — status messages, null states, first-run patterns
• IBM Carbon Design System — action labels, writing style
• IBM Carbon for AI — AI labeling principles
• Mailchimp Voice and Tone — voice/tone framework, grammar, word list
• Nielsen Norman Group — empirical research on errors, empty states, GenAI UX

Plus principles from Apple HIG — Writing, Material Design 3 — Content Design, Atlassian Design System — Voice and Tone, Torrey Podmajersky's Strategic Writing for UX (2nd ed., 2024), and Kinneret Yifrah's Microcopy: The Complete Guide (2nd ed.).

When to invoke

The skill triggers when you ask for:

• "Write a [button label / error message / empty state / onboarding copy] for X"
• "Audit this UI copy"
• "Rename this feature / setting / section"
• "Write copy for an AI suggestion / loading state / failure"
• "Set up a voice & tone for our product"
• "What should this error say when X fails?"

It will not trigger on marketing copy requests — and will redirect if you ask for one.

What you get back

For writing tasks: 2–3 options with one-line rationale each, ordered by recommendation, plus flagged risks (localization, accessibility, voice mismatches, blocklist words).

For audits: each finding names the issue, quotes the current copy, proposes a fix, and explains why the fix is better in one sentence.

License

MIT — see LICENSE.

Contributing

PRs welcome, especially:

• Additional verbatim sourced examples from US design systems (the AI content, settings/toggles, and form labels sections have flagged gaps)
• Updates when sources update (e.g., when Apple HIG / Carbon for AI publish more do/don't guidance for AI content)
• Translations of the rules into other languages (the principles are language-agnostic; some pattern guidance is English-specific and would need adaptation)

Please cite sources with deep-link URLs for any new examples.