atomise

Current Providers

• [skills.sh](https://skills.sh) - Community-curated skills leaderboard

- All Time (/) - Total installs ranking

- Trending (/trending) - Recent growth ranking

- Hot (/hot) - Daily installs ranking

Planned Providers

• GitHub Trending - Popular skill repos on GitHub
• Awesome Lists - Curated awesome-* lists for AI agent skills

Manual Skills

Skills not tracked by any provider can be manually added via data/manual_skills.json:

```json

{

"skills": [

{

"source": "owner/repo",

"skillId": "skill-name",

"name": "Skill Display Name",

"installs": 1

}

]

}

```

Manual skills will be:

• Fetched for their SKILL.md from GitHub (using standard skill folder detection)
• Included in skills_index.json with providerId: "manual"
• Not overwritten by the crawler (they persist across runs)
• Deduplicated: If skills.sh later tracks a manual skill, it will use skills.sh data instead

Note: installs should be at least 1 (minimum value).

The crawler generates files in the data/ directory:

`data/skills.json`

Complete skill data containing all three leaderboards:

```json

{

"updatedAt": "2024-01-27T00:00:00.000Z",

"allTime": [...],

"trending": [...],

"hot": [...]

}

```

`data/skills_index.json`

Website-friendly index for all skills (built from data/skills.json):

• Includes description as a path to description_en.txt (when a cached SKILL.md exists under data/skills-md/)
• Includes skillMdPath so your website can fetch and render the full markdown
• Deduplicated by id (/). If the upstream data contains duplicates, the index keeps the entry with the highest installsAllTime.

`data/feed.json`

Simplified feed format (top 50 from each leaderboard).

It also tries to enrich each item with a description by fetching the corresponding GitHub SKILL.md (cached under data/skills-md/):

```json

{

"title": "Skills Feed",

"description": "Aggregated AI agent skills from multiple sources",

"link": "https://github.com/user/skills_feed",

"updatedAt": "2024-01-27T00:00:00.000Z",

"topAllTime": [...],

"topTrending": [...],

"topHot": [...]

}

```

`data/skills-md/`

Cached SKILL.md files fetched from GitHub, using common skill folder locations such as:

• skills//SKILL.md (most common)
• .claude/skills//SKILL.md
• .cursor/skills//SKILL.md
• .codex/skills//SKILL.md
• plugins//skills//SKILL.md (common in plugin-based repos, e.g. Expo)

When a SKILL.md is present, the crawler also generates:

• description_en.txt (extracted from the SKILL.md frontmatter description when available)

By default, the crawler only fetches SKILL.md for skills included in the top lists (to keep the daily job fast).

If you really want to sync all skills from data/skills.json, you can run:

```bash

SYNC_ALL_SKILL_MDS=1 bun run crawl

```

`data/feed.xml`

RSS 2.0 feed (XML). This is meant for RSS readers / subscriptions.

• It is generated from the current crawl + the previous data/feed.json
• It only publishes meaningful changes (new entries / rank jumps) to avoid spamming

Local Development

```bash

# Install dependencies

bun install

# Run crawler

bun run crawl

```

Tip: if you want more complete GitHub SKILL.md coverage (including plugin-style paths like plugins/*/skills/...),

set GITHUB_TOKEN to avoid GitHub API rate limits:

```bash

export GITHUB_TOKEN=ghp_xxx

bun run crawl

```

GitHub Actions

After pushing to GitHub, the crawler will:

1. Run automatically daily at UTC 0:00
2. Support manual triggering (click "Run workflow" in Actions tab)
3. Run automatically on push to main branch

You can fetch data directly via GitHub Raw URL:

```

https://raw.githubusercontent.com///main/data/skills.json

```

Or use jsDelivr CDN (faster):

```

https://cdn.jsdelivr.net/gh//@main/data/skills.json

```

RSS Subscription (Recommended)

Subscribe to the RSS feed:

```

https://raw.githubusercontent.com///main/data/feed.xml

```

Or via jsDelivr CDN:

```

https://cdn.jsdelivr.net/gh//@main/data/feed.xml

```

Example Code

```typescript

// In Next.js

const SKILLS_DATA_URL = 'https://cdn.jsdelivr.net/gh/your-username/skills-crawler@main/data/skills.json';

export async function getSkillsData() {

const res = await fetch(SKILLS_DATA_URL, {

next: { revalidate: 3600 } // Revalidate every hour

});

return res.json();

}

```

• Data is updated daily
• Please comply with each provider's terms of service
• For personal learning and research purposes only

Want to add a new skill source? PRs are welcome! Check out the existing provider implementations in the codebase.