An open registry of Claude Code Skills — community-contributed SKILL.md files for Anthropic's Claude Code agent. Updated twice daily from 26 public sources. No author pays for inclusion or ranking.
For the full experience including multi-signal Quality Score and one-click install features for each skill — upgrade to Pro.
The promise: ClaudSkills is an objective discovery layer for Claude Code Skills. No skill author pays anything — no listing fee, no acceleration, no ranking boost, no featured placement, no sponsored Pro Collections, no "verified author" upsell, no affiliate kickback from external directories. The catalog is mined automatically; admission is rule-based; ranking (when Pro is enabled) is derived from content signals only.
SKILL.md file is scored on structural quality and admitted if it clears a single fixed threshold. No human curation in the admission path.site/data/attribution_overrides.json./data/skills.json. Per-skill markdown is available at /skills/<slug>/SKILL.md.optout@claudskills.com; verified requests are processed within 24 hours and propagated to the next miner cycle.| Category | Skills | Browse |
|---|---|---|
| General | 36,000+ | /category/general/ |
| Engineering | 13,000+ | /category/engineering/ |
| Security | 3,500+ | /category/security/ |
| Dev Tools | 3,000+ | /category/tools/ |
| Content | 3,000+ | /category/content/ |
| Science & Research | 2,500+ | /category/science/ |
| Product | 2,500+ | /category/product/ |
| Growth | 1,500+ | /category/growth/ |
| Sales | 1,200+ | /category/sales/ |
| Advertising | 200+ | /category/ads/ |
Counts are rounded to the nearest hundred so the page stays stable between miner cycles. Live counts are derived from /data/skills.json on every refresh.
Sources span code hosts (GitHub, GitLab, Codeberg, HuggingFace), aggregators (awesome lists, marketplaces, RSS feeds), and discussion surfaces (Reddit, HackerNews, dev.to, Mastodon, Bluesky, Telegram, YouTube). Each source is rate-limited and wrapped in a fail-soft section so one broken endpoint cannot kill the run.
SKILL.md files — gists + raw markdownFull source code at scripts/skill_miner.py. Every source has its own admission counter, hash-deduplication against prior runs, and per-source budget cap.
Full catalog as a single JSON file. One object per skill — slug, name, description, category, sub-category, tags. Refreshed every miner cycle. CORS-open for client-side use.
Individual SKILL.md markdown files served via Cloudflare R2 for read-only mirroring at scale. Identical to the version installed when a developer runs claudskills install <slug>.
GET https://claudskills.com/skills/<slug>/SKILL.md
Three sub-sitemaps referenced from the index — pages, skills, authors. Standard <urlset> format with lastmod and priority.
Last-100-skills feed in three formats served via a Cloudflare Worker, with WebSub hub announcement for low-latency aggregator discovery. See the promotion-engine action plan for status.
Full catalog mirrored as a HuggingFace dataset (Parquet + JSON), refreshed daily. Tagged for LLM training pipeline discovery.
Read-only repository at github.com/claudskills/catalog-public. Same data as the JSON above, with per-row attribution + a CITATION.cff for academic citation.
Free browsing exposes the full unranked catalog. Pro subscribers see a multi-signal QualityScore on every skill in the desktop app, computed server-side from content structure and frontmatter depth. The score is rescaled to a stable [50, 100] reference band so adding new skills never reshuffles existing scores. See the about page for the methodology in full.
If you reference ClaudSkills data in a paper, blog post, or library: please cite ClaudSkills with the URL https://claudskills.com/stats/ and the access date. A CITATION.cff entry will land alongside the public catalog mirror once Phase 2 of the promotion engine ships.
SKILL.md files, not affiliated with, endorsed by, or sponsored by Anthropic.