Agent Skills Board

Agent Skills Board is a crawler-first registry and local dashboard for discovering public Agent Skills—folders built around SKILL.md on GitHub—then indexing them for search, filters, scoring, and detail views (data/skills.json + API + static UI).

Getting started

export GITHUB_TOKEN=ghp_your_token_here   # https://github.com/settings/tokens
bash run.sh

Open http://localhost:8000

For why the project exists, scope boundaries, and north-star direction, see VISION.md.

Ecosystem

Resource	Role
Agent Skills (spec & docs)	Open format maintained by Anthropic and the community—what a skill is.
agentskills.io	Documentation and tutorials for the format.
Agent Skills Board (this repo)	Discovery layer—finds SKILL.md in the wild, normalizes metadata, scores signals, and serves API + static UI.

What’s in this repository

• Crawler — GitHub code search, repo trees, and raw fetches with ETag-friendly caching.
• Normalizer — Front matter parsing, role hints, and transparent scoring.
• Index — data/skills.json (and raw crawl output) as a portable, diffable artifact.
• API + dashboard — FastAPI backend and a zero-build-step HTML/CSS/JS UI.

How it fits together

GitHub API (code search + repo tree)
        ↓
  crawler.py          ← finds SKILL.md paths, fetches content
        ↓
  normalizer.py       ← parses frontmatter, scores skills
        ↓
  data/skills.json    ← flat index
        ↓
  api.py              ← search, filter, stats
        ↓
  dashboard/          ← static UI

API usage (summary)

Phase	Strategy	API calls (order of magnitude)
Pinned repos	One tree per repo	~10
Official orgs	One code search per org	~20
Broad search	Paginated code search	a few pages
Content	raw.githubusercontent.com	no REST quota
Re-crawl	ETag → 304 saves quota	—

With GITHUB_TOKEN: 5,000 REST requests/hour; a full crawl is typically on the order of hundreds of calls. Without a token, stay within unauthenticated limits (for example tune RATE_LIMIT_DELAY in config.py).

Scoring (at a glance)

Signals include official-org allowlists, stars (log-scaled), package downloads where available, recency, and light CI/tests heuristics. Weights live in config.py (SCORE_WEIGHTS) so you can tune discovery for your org.

Repository layout

agentskillsboard/
├── config.py           ← org lists, score weights, crawler settings
├── crawler.py
├── normalizer.py
├── api.py
├── run.sh
├── requirements.txt
├── VISION.md           ← intent, scope, roadmap
├── data/
│   ├── raw_skills.json
│   └── skills.json
└── dashboard/static/index.html

Extending the board

Pinned repos, org allowlists, and score weights are the main knobs today. To index other capability shapes (for example MCP server directories), treat them as explicit schema and scope—see the “federation” notes in VISION.md rather than bolting on undocumented scrapers.

---

Agent Skills Board focuses on transparent, local-first discovery for one family of artifacts (SKILL.md skills). The agentskills repository defines the portable format those artifacts follow.