About AgentSkillsHub

An open-source directory for Claude Skills, MCP Servers & Agent Tools — built in public, scored transparently.

Who runs this

AgentSkillsHub is built and maintained by Jason Zhu ( on X), an independent researcher tracking open-source AI agent ecosystems since 2024.

Jason is also the author of the Blue Book of Agent Skills 2026 — an in-depth analysis of the Claude Skills / MCP / Codex ecosystem based on 62,000+ indexed repositories, published openly on GitHub with reproducible Python scripts.

The Hub is a solo project without external funding. There is no sales team, no "pay for placement" tier, and no dark patterns. All editorial decisions and scoring formulas live in public on GitHub under MIT license.

What we track

62,000+
indexed skills
7
categories
58
scenario pages
8h
refresh cadence
10
quality dimensions
MIT
source license

Every GitHub repository matching our search patterns flows through a 6-phase pipeline every 8 hours: search → metadata enrichment → README fetch → quality scoring → category classification → composability analysis. No manual curation is applied to ranking — all signals are data-driven to avoid editorial bias.

Editorial methodology

The 10-dimension quality score

Every skill receives a composite quality score (0-100) based on ten weighted signals. The formula and weights are fully documented in services/scorer.py:

  • Completeness (15%) — README depth, license, description, stars
  • Clarity (15%) — description quality, topic tags, naming conventions
  • Specificity (15%) — language + topic count + category + size
  • Examples (12%) — code samples, commit frequency, contributor count
  • README structure (23%) — sections, code blocks, badges, TOC
  • Agent readiness (20%) — SKILL.md presence, install command, MCP compliance

On top of the quality score, a separate composite score weighs nine additional signals including stars, recency, forks, commit velocity, issue resolution rate, star momentum (Z-score normalized), author followers, and a size bonus favoring atomic skills. See the full methodology.

Daily & weekly curation

Daily reports feature skills newly indexed in the last 48 hours with at least 20 GitHub stars. The weekly "Trending" list ranks by star_velocity (stars per day over the last 7 days). Both are automated but the final Top 10 is human-reviewed to exclude obvious forks, batch spam accounts, and misclassified non-skills.

How decisions are made transparently

Editorial judgement is required in exactly three places:

  1. Category classification corrections — when keyword inference misfires
  2. Spam & batch-farming exclusion — documented in discover_candidates.py
  3. Verified Creator decisions (Q2 2026 launch) — criteria and outcomes logged publicly

Every other ranking decision is purely data-driven and reproducible from the open-source codebase.

Source code & data

Every line of code and every data snapshot used for the Blue Book is public:

Contact

Questions, corrections, or tips? Reach out via:

← Back to Explore All Skills