metadata-check — Claude Code skill: metadata review
metadata-check reviews short text such as titles, meta descriptions, FAQs, bios, and taglines. Use it when each field must stay clear, concrete, and within its target length.
MIT 171 installs ★ 110 updated 2026-07-05
npx skills add jdevalk/skills --skill metadata-check -a claude-codeWhat metadata-check does
This skill audits short strings that do not fit ordinary readability checks: page titles, meta descriptions, schema descriptions, FAQ answers, GitHub repository taglines, profile bios, and social-card copy. It checks whether the searchable term comes early, the wording is concrete, filler is removed, the voice is active, and the description adds information instead of repeating the title.
For every string, the skill reports separate checks and supplies a replacement when a problem is found. It also reviews difficult words, search-result truncation, and the one-idea-per-field rule. Route multi-paragraph prose to readability-check instead.
In Claude Code, installation uses npx; the skill file is then placed at ~/.claude/skills/metadata-check/SKILL.md. The scope is narrow: URLs, filenames, and other technical identifiers are skipped.
What it helps with
- audit a page title before publishing
- fit a meta description to search-result length
- spot title and description repetition
- rewrite an FAQ answer in active voice
- remove filler and difficult wording from a tagline
Example prompts for the agent
- Review this page title and meta description against every skill check.
- Audit these FAQ answers and rewrite each string that fails.
- Check this GitHub repository tagline and profile bio for length, concrete wording, and one idea per field.
How to install metadata-check
1Skills CLI (Vercel)
npx skills add jdevalk/skills --skill metadata-check -a claude-code2Manually: put the files in place
~/.claude/skills/metadata-check/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/metadata-check/
1Skills CLI (Vercel)
npx skills add jdevalk/skills --skill metadata-check -a codex2Manually: put the files in place
.agents/skills/metadata-check/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add jdevalk/skills --skill metadata-check -a cursor2Manually: put the files in place
.cursor/skills/metadata-check/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install jdevalk/skills metadata-check2Manually: put the files in place
.github/skills/metadata-check/SKILL.md1Skills CLI (Vercel)
npx skills add jdevalk/skills --skill metadata-check -a gemini-cli2Platform CLI
gemini skills install https://github.com/jdevalk/skills --path metadata-checkManually: put the files in place
.devin/skills/metadata-check/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add jdevalk/skills --skill metadata-check -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about metadata-check
How do I install metadata-check in Claude Code?
Run npx skills add jdevalk/skills --skill metadata-check -a claude-code. The skill file will be at ~/.claude/skills/metadata-check/SKILL.md.
What does metadata-check do?
It audits short copy for concrete wording, active voice, length, repetition, and readability. When a string fails a check, it gives a ready-to-use rewrite.
What can metadata-check review?
It handles titles, meta descriptions, schema descriptions, FAQ answers, GitHub taglines, bios, and social-card copy. Use readability-check for multi-paragraph prose.
Source text
SKILL.md · MIT
--- name: metadata-check description: > Reviews short high-value strings — page titles, meta descriptions, schema description fields, FAQ answers, GitHub repo taglines, profile bios, social-card copy, and other metadata where Flesch and paragraph-level readability checks don't apply. Checks front-loading, concreteness, filler, active voice, title/description duplication, difficult words, SERP-truncation fit, and one-idea-per-field. Use when the user asks to review metadata, a tagline, a bio, or SEO strings, or proactively after generating any short string with audience-facing impact. Chained into by astro-seo, wp-readme-optimizer, github-repo, and github-profile for their metadata outputs. For multi-paragraph prose, use the `readability-check` skill instead. --- # Metadata check Reviews short, high-value strings where every character counts — titles, meta descriptions, schema descriptions, FAQ answers, taglines, bios, and social-card copy. Flesch scoring and the ten-category readability rubric don't fit a 5–30 word string; this skill applies the checks that do. Use this skill for anything in the metadata-string shape. For multi-paragraph prose (blog posts, READMEs, CONTRIBUTING files), use `readability-check` instead. ## What to check For each string, check these points. Every check is either ✓ (pass), ⚠ (needs work), or ✗ (problem). - **Front-load the distinguishing word.** The most specific, searchable term sits near the start. "Astro SEO: the definitive guide" beats "The definitive guide to Astro SEO" — the reader's eye hits the topic first. - **Concrete over abstract.** Names, numbers, specific claims. "Reduce LCP by 40%" beats "Improve performance significantly." - **Filler and hedging — aggressive.** Every word costs shelf space. Cut *really*, *just*, *very*, *actually*, *basically*, *simply*, *a bit*, *kind of*, *I think*. In metadata these aren't style ticks, they're wasted SERP characters. - **Active voice unless the object is the subject.** "Audit your GitHub profile" beats "Your GitHub profile can be audited." - **No title/description duplication.** A meta description that restates the title burns ~60 of 200 characters for nothing. The description should promise what's *inside* the page, not re-announce what it *is*. - **Difficult words.** Prefer the common synonym unless the domain term is what the user searches for. Between *utilize* and *use*, use *use*. Domain terms the target audience expects are fine (*hydration*, *middleware*, *structured data*). - **SERP and platform truncation.** Know the limit for the surface: - Google SERP: titles clipped around 55–65 characters, descriptions around 155–160. - GitHub repo description: ~350 chars, but only the first ~100 show in search results and on profile cards. - Twitter/X bio: 160 chars. - Open Graph description: usually shown at ~200 chars. Anything past the cut should be disposable, not a key promise. - **One idea per field.** Titles and descriptions that try to promise two things land fuzzy. Pick the sharpest one. ## What to skip Don't run this skill on: - URLs, schema `@id` values, filenames, enum values, or any technical identifier. - Multi-paragraph prose — route that to `readability-check`. - Strings the user did not write and cannot change (imported content, upstream defaults). ## Output format Per string, return: ```markdown ### [Field name] — [length] chars > [The string itself] - Front-loading: [✓/⚠/✗] [one-line reason if not ✓] - Concreteness: [✓/⚠/✗] [reason] - Filler: [✓/⚠/✗] [reason] - Active voice: [✓/⚠/✗] [reason] - Duplication: [✓/⚠/✗] [reason] - Word choice: [✓/⚠/✗] [reason] - Truncation fit: [✓/⚠/✗] [reason] — platform: [SERP/GitHub/Twitter/OG] - One idea: [✓/⚠/✗] [reason] **Rewrite:** [only if any check failed — a concrete replacement string, not a suggestion] ``` Keep the audit tight — metadata feedback should be actionable in seconds, and you're almost always running it across a batch of strings at once. If a string is already clean, don't pad the output; just mark all ✓ and move on.
© jdevalk · MIT · jdevalk/skills