How ClaudeAtlas Works

Discovery

ClaudeAtlas automatically discovers skills by scanning GitHub for SKILL.md files using the GitHub Code Search API. We use size-range partitioning to exceed the API's 1,000-result cap, supplemented by GitHub Topics search and known skill repositories. The scraper runs daily via GitHub Actions and has discovered 42,321 skill files so far — of which 1,286 pass our filters and are indexed, with 361 earning the Featured tier.

Quality Scoring

Every discovered skill is scored on a 100-point scale using 7 signals, all sourced from the GitHub API. The formula is fully transparent:

Signal Weight What It Measures
GitHub Stars 20% Community trust (log-scaled to avoid mega-repo bias)
Last Commit 20% Freshness — skills in a fast-moving ecosystem need maintenance
Frontmatter 20% SKILL.md completeness — has name, description, and metadata
Documentation 15% Skill body length + repo description quality
Issue Health 10% Open issues relative to popularity — high ratio suggests abandonment
License 10% Open-source license presence signals professional quality
Description 5% Repository has a meaningful description

Tiers

Skills are assigned to three tiers based on their composite score:

Featured Score 80+ — Top-quality skills prominently displayed
Solid Score 50-79 — Good skills that meet quality standards
Listed Score below 50 — Meets minimum bar but not highlighted

Filtering

We automatically exclude archived repositories, forks, and repos with no stars. Skills from deleted or private repos are removed after 3 consecutive days of returning 404. Repos not updated in over 6 months receive a recency penalty.

Categories

Skills are auto-categorized into 8 categories using keyword matching on the skill name, description, and repository topics. The mapping is deterministic and consistent across runs.

Freshness

The entire index is re-scraped and rebuilt daily via GitHub Actions. ETag caching ensures efficient API usage — unchanged repos are skipped automatically. The "Updated daily" claim is literal, not aspirational.

Open Data

The scored skills data is available as JSON. The scoring algorithm is open source. If you believe a skill is miscategorized or mislabeled, please open an issue on GitHub.