Suede AI Skills Directory · Data

API and data

Everything the directory publishes is a static file on this host. There is no key, no rate limit and no account; please link back to the listing when you quote a score, and name skills.sh when you quote an install count.

Files

FileURLWhat it holds
Search index/search-index.jsonOne record per listed skill (3,871): name, repo, description, category, license, grade, score, href, full, report, lane, flags, topics, listed, updated, installs.
Skill report/activeloopai/hivemind/hivemind-goals/report.jsonA verifiable report per listing: snapshot id and dataset hash, score by component, source revision, badges and the methodology version. Every listed skill has one at its listing path.
Quality badges/activeloopai/hivemind/hivemind-goals/score.svgscore.svg and badge.svg per listing. Skills scoring 80 to 99 get the numeric and letter shields; others serve the general listing badge.
Install snapshot/installs.jsonInstall counts for 2,869 listed skills as published by skills.sh, with the fetch date and the method. The directory measures no installs of its own.
Kinds/kinds.jsonThe plugins, Grok, MCP server and rules listings as one JSON document.
Finder skill/finder/Install the read-only directory search skill for Claude Code, Codex or Cursor.
Weekly feed/weekly.jsonRising and climbing skills, topic growth, new listings and repository stars for the latest snapshots; read by the finder’s weekly briefing. Suede AI excluded.
State of Agent Skills/state-of-agent-skills/data.jsonSnapshot-based grade distribution, category and license counts, and score-component gaps; Suede AI excluded.
RSS feed/feed.xmlNew listings and guides, RSS 2.0.
llms.txt/llms.txtThe directory described for answer engines, with every browse page.
Sitemap/sitemap.xmlEvery indexable page with its last-modified date.

Where install counts come from

The directory does not measure installs. installs.json is a snapshot of counts published by skills.sh, which records installs made with the skills CLI (npx skills add) from anonymous, deduplicated telemetry as its documentation describes. The snapshot was read on November 8, 2026 from its public search endpoint, one query per listed repository, and its home page leaderboard for eight weeks of history. 2,869 listed skills matched; skills without a count are not tracked there. The script is scripts/directory/installs.mjs.

Recomputing a score

Scores are computed only from published fields, so a report can be checked. scripts/directory/badges.mjs and scripts/directory/grade.py in the repository recompute the same value from skills.json; the methodology lists every component and its points.