Public API

Quick start

Every endpoint is GET-only, JSON, no auth, CORS open. Hit any URL straight from a browser, curl, or an LLM tool-call:

curl https://claudskills.com/api/v1/skills?limit=5
curl https://claudskills.com/api/v1/skills/seo
curl https://claudskills.com/api/v1/categories
curl https://claudskills.com/api/v1/openapi.json

Or in JavaScript:

const r = await fetch('https://claudskills.com/api/v1/skills?category=engineering&limit=20');
const {data, next, total} = await r.json();
// data: array of Skill objects; next: URL for the next page (null if last)

Endpoints

Skill schema

Every Skill object in API responses has this stable shape:

{
  "slug":           "seo",
  "name":           "SEO",
  "description":    "Run a broad SEO audit across…",
  "category":       "general",
  "subcategory":    "general-misc",
  "tags":           ["type:audit"],
  "daily_eligible": false,
  "featured":       false,
  "author":         null,
  "author_url":     null,
  "license":        null,
  "url":            "https://claudskills.com/skills/seo/",
  "og_image":       "https://claudskills.com/og/seo.png"
}

Fields will only be added in v1 — never removed or renamed. Breaking changes will mount at /api/v2/ in parallel with v1 still running.

Caching & rate limits

There is no enforced per-key rate limit. The Worker is fronted by Cloudflare's edge cache with a 1-hour TTL, so even high-volume callers hit the upstream catalog rarely. The catalog itself only changes daily (01:00 local miner run; typically settled by ~05:00 local).

Every response carries Cache-Control: public, max-age=3600, s-maxage=3600. Please respect it on your end — repeating identical requests inside the TTL window is wasted bandwidth.

If you need to enumerate the full catalog (all ~69,000 skills), prefer the bulk dataset at github.com/claudskills/catalog-public — one daily-refreshed file in JSON, NDJSON, CSV, and Parquet formats, plus an Atom feed of recent additions. Cheaper, faster, and Git-versioned.

Related surfaces

The API is one of four programmatic surfaces over the same catalog:

• /api/v1/ — this REST API (what you're reading)
• catalog-public — daily bulk dataset (JSON, NDJSON, CSV, Parquet, Atom). CC BY 4.0. Best for full-catalog ingest.
• HuggingFace dataset — same data, parquet-native, suitable for LLM training and dataset tooling.
• /llms.txt — flat-text index optimised for LLM crawl ingestion. Top 500 indexed skills + tracked LLM citation queries.

Plus two browsing-flavoured embed surfaces:

• /embed/<slug>.js — one-line <script> snippet that injects a styled skill card into any blog post or docs page.
• /badge/<slug>.svg — shields.io-style SVG badge for skill authors to drop into their GitHub README.

License & attribution

The catalog data (what the API returns) is licensed CC BY 4.0 — free for any use including commercial, attribution required. The minimal attribution is a link to claudskills.com somewhere near where the data is displayed.

The per-skill SKILL.md content is the property of the original author and carries whatever license the author chose. Each Skill object's license field reflects what the upstream repo declares. If a skill's license is null we couldn't determine it — treat as restrictive until you check the upstream repository.

We do not require notification of API use, registration, or revenue share. If you build something interesting on top of the API, we'd love to hear about it: [email protected].

Versioning

The API is mounted at /api/v1/ with a stable contract — fields will only be added, never removed or renamed within v1. Breaking changes will mount at /api/v2/ in parallel. We will announce v2 at least 6 months before sunsetting v1, with a clear migration guide and dual-running period.

FAQ