AGENT SKILL

cratis-discoverability

Check Cratis content before publishing for search and AI answers without sacrificing readability. Use when reviewing discoverability, SEO, AEO or GEO; writing fact blocks or download claims; editing package/repo metadata; checking docs titles, descriptions or LLM export budgets; or reviewing founder/company JSON-LD and member profile indexing. Not for page voice and structure (cratis-documentation-writing), llms.txt design (cratis-llm-friendly-documentation), distribution plans (cratis-developer-marketing) or length decisions (cratis-content-length).

License 许可
MIT
最近更新
2026-11-14
Stars
★ 0
Forks
0

01 前言 Front Matter

name
cratis-discoverability
license
MIT
description
Check Cratis content before publishing for search and AI answers without sacrificing readability. Use when reviewing discoverability, SEO, AEO or GEO; writing fact blocks or download claims; editing package/repo metadata; checking docs titles, descriptions or LLM export budgets; or reviewing founder/company JSON-LD and member profile indexing. Not for page voice and structure (cratis-documentation-writing), llms.txt design (cratis-llm-friendly-documentation), distribution plans (cratis-developer-marketing) or length decisions (cratis-content-length).

02 资源文件 7

03 SKILL.md 正文

<!-- cratis-ai-managed: skills/cratis-discoverability/SKILL.md -->

Discoverability before you publish

Discoverability is a property of good content: a clear definition, an honest title, one name per thing and working links. The always-on rule is discoverability.md; this skill is the procedure to run on the finished draft. If a fix makes the page worse to read, drop the fix and say so.

Procedure

  1. Name the content type: docs page, blog post, landing page, README, package/repo metadata or member profile.
  2. Name the reader's one job and, for docs, the page type (tutorial, how-to, reference, explanation).
  3. Apply the checks for that type below, including facts, adoption claims and privacy without waiting for a separate request. Fix the draft; do not annotate it.
  4. For a built site, run the checker and fix errors; warnings are judgment calls to report.
  5. Report what was changed, what was left alone and why.

Read references/page-checklist.md when checking a page, including docs lint baselines and LLM export budgets.

Checks by content type

  • Docs page. The description is written by hand as one self-contained definition or outcome sentence. Tutorials, getting-started pages and explanations follow writing-cratis-docs (scenario first); the definition lives in the description, and one definition sentence may follow the opening scenario where it reads naturally. writing-cratis-docs wins on voice and structure. Product index pages open with the definition sentence. Reference pages stay terse: title is the exact name plus its product, no question headings, no intro. Titles are unique across the site and name the product ("Chronicle projections"). Keep description and duplicate-title lint baselines at 0; never raise a baseline to pass. Check rendered titles and measure llms.txt / llms-small.txt bytes against the site budgets.
  • Blog post. The title states the topic in the words a reader would search. The intro gives the conclusion in two sentences. Show the named author and date, add lastUpdated only for real changes, link the owning docs pages with https://www.cratis.io/..., keep description apart from the card excerpt.
  • Landing page. One primary term in the title, the <h1> or the first paragraph, and a one-sentence definition above the fold. Brand lines stay; pair them with a category term in the title and description. Respect the claim gate in BRAND.md, MESSAGING.md and SITE.md.
  • README. The H1 may be the product name ("Cratis <Product>"). The first sentence defines the product in 15-35 plain nouns, with no logo-only or metaphor opener. Links are absolute; link the docs page.
  • Package metadata. One sentence naming what the package is and which Cratis product it belongs to, identical to the README opener. PackageProjectUrl and homepage point at the docs page, not only at GitHub. Read references/package-metadata.md when editing package fields, GitHub About, Docker Hub descriptions, analyzer help links, template messages or README footers. Check these surfaces together; Authors and Company are Cratis.

Fact blocks and adoption claims

For each entity page, assemble one sourced "at a glance" block from verifiable facts, with dates for counts. Read references/facts-and-adoption.md when selecting facts or calculating downloads: identify entry packages and continuation predecessors, exclude sub-packages and overlapping shared dependencies, round down to a stated threshold, and publish the month and registries. Downloads are never users or customers. Omit unsupported claims instead of inventing customers, user counts, benchmarks, certifications, prices or competitor comparisons. Keep shared guidance generic; do not copy private integration names, legal entity details or factual counts into it.

Names

Use the canonical names, one per concept, and the first-mention rule. Read references/entities.md when naming a product, writing a definition sentence or filling sameAs. Do not invent a name, URL, maturity word or claim that the references do not state.

Structured data

Describe only what is visible on the page, one @graph per page, stable @ids. Add no HowTo and no new FAQPage. Read references/structured-data.md for the patterns and what each page type gets. Use only public, owner-approved founder/company facts. Do not name other people in structured data; never add personal data or Person entities for community members. Member profiles are noindex; check profile and CV routes, rendered robots directives and JSON-LD. Without approved public founder facts, reference the organization by @id instead.

Checking a built site

For a built static site (Astro dist/ or a plain-HTML repository root), run with Node.js: node <skill-directory>/scripts/verify-discoverability.mjs --root <dir> --site <https://host>. It reports <html lang>, title, description, canonical host, <h1> count, JSON-LD parse errors, robots.txt, the sitemap and llms.txt structure. Errors exit 1; warnings do not. Pass --warn-only to downgrade errors and --require-llms where the site ships llms.txt. Read scripts/verify-discoverability.mjs only when changing the checker.

Pitfalls

  • Question headings everywhere: use them only where they read naturally (concept pages, FAQs, comparisons, blog), never on reference pages or tutorial steps.
  • Padding to a word count: there is no minimum length; a complete page is done.
  • Apex https://cratis.io/ links: the docs host is https://www.cratis.io.
  • A description that repeats the title or a sibling page's description.

04 同仓库更多 Skills