The free, open marketplace of
.mdskill files for Claude Projects.
Download once. Claude applies it automatically in every conversation. Forever.
Every time you start a new Claude conversation, you re-explain the same instructions. Copy-paste. Again and again.
SkillGalaxy fixes that.
It's a community-powered library of .md skill files designed for Claude Projects. Add a skill once to your project instructions — Claude applies it automatically in every conversation from that point forward.
No repeated prompts. No wasted tokens. Just results.
1. Browse SkillGalaxy → find a skill
2. Click Download .md
3. Open Claude → go to a Project → paste into Project Instructions
4. Done. Claude applies the skill in every conversation automatically.
No account needed to download. No friction.
| Skills | 10,171+ and growing |
| Domains | 16 |
| Price | Free. Always. |
| Model | Community-built, reviewed within 24hrs |
AI & ML · Cybersecurity · Data Engineering · Cloud & Infra · Quantum Computing · Computational Biology · Spatial Computing · Blockchain & Web3 · Robotics & Automation · Climate Tech · Product & Strategy · Creative Technology · Development · Writing · Business · Design & Education
Every skill now has a 1–5 star rating widget. Open any skill → scroll to Rate this Skill → leave a rating and optional review. Average scores and recent reviews load dynamically from Supabase.
Community submissions are automatically scored by Claude (claude-haiku) on submit. Skills receive a quality score (1–10) with actionable feedback. Requires ANTHROPIC_API_KEY in your Vercel environment — see AI Validator Setup.
Click ** Connect Claude Desktop** in the hero to get a ready-to-copy claude_desktop_config.json snippet that wires the SkillGalaxy MCP server directly into Claude Desktop.
Skills now carry a version (semver) and changelog field. Contributors can bump the version when they update a skill — users see "Updated X days ago" and the version in the skill detail modal.
The most advanced skill composition engine in the ecosystem. Instead of naively concatenating skills, SkillForge:
- Builds a synergy graph — models all skills as nodes, edges represent shared tags, category alignment, and proven composition synergies
- Runs evolutionary optimization — genetic algorithm (selection → crossover → mutation → pruning) discovers optimal skill subgraphs
- Prunes overlapping content — detects and removes redundant instructions between composed skills
- Generates forged composites — outputs a single
.mdfile with merged context, provenance tracking, and token budget
Benchmark results (100 skills, 50 queries):
| Metric | Baseline (Top-K) | SkillForge | Delta |
|---|---|---|---|
| Avg Tokens | ~4,800 | ~3,200 | -33% |
| Success Rate | 0.78 | 0.89 | +14pp |
| Latency | O(N) | O(G×P×E) | <500ms |
Access via:
- Frontend: Click "⚡ Forge Composer" in the sidebar
- API:
POST /api/skillforge-composewith{ query, skillSummaries } - MCP:
forge_compositetool (10th tool in MCP v2.0)
# Run the benchmark simulation
python3 scripts/skillforge_benchmark.pyConnect Claude directly to the entire SkillGalaxy library via Model Context Protocol. No manual downloads needed.
cd mcp-server && npm install && npm run build-skillsAdd to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"skillgalaxy": {
"command": "node",
"args": ["/path/to/skill_galaxy/mcp-server/index.js"]
}
}
}Restart Claude Desktop — done. Claude can now search, compose, visualise, and orchestrate all 10,171+ skills.
npm run start:http # http://localhost:3100/mcp
MCP_API_KEY=secret npm run start:http # with bearer auth| Tool | Description |
|---|---|
search_skills |
Search by keyword/category/difficulty (L1 metadata only) |
get_skill |
Load full markdown body (L2) |
get_skill_deep |
Full body + related skills + cross-references (L3) |
get_skill_summary |
Compact metadata table |
list_categories |
Browse all 16 categories |
execute_skill_code |
Code Mode — run JS pipelines server-side (~37% token savings) |
compose_skills |
Chain skills into workflows with reflection |
generate_visual |
Mermaid diagrams, comparison tables, dashboards |
ingest_file |
Process PDFs, CSVs, XLSX, images natively |
forge_composite |
SkillForge — evolutionary graph optimization for optimal skill compositions |
- Progressive Disclosure — 3-level loading keeps context minimal
- Code Mode — orchestrate multiple tools in one call, reducing token usage
- Composable Workflows — chain skills with optional self-reflection
- Generative UI — Mermaid diagrams, learning paths, category dashboards
- Security — input sanitisation, rate limiting, audit logging
- Binary Ingestion — PDF, CSV, XLSX, image processing
- SkillForge — evolutionary skill graph composition with token pruning
- Streamable HTTP — remote access with bearer auth
The AI validator scores submitted skills automatically using Claude Haiku.
- Get an Anthropic API key
- In your Vercel project → Settings → Environment Variables → add:
ANTHROPIC_API_KEY = sk-ant-... - Redeploy. Done. The
/api/validate-skillendpoint becomes active.
Without the key the validator is silently skipped — submissions still work normally.
After cloning and running supabase-setup.sql, run the migration to enable ratings, reviews, and versioning:
-- In Supabase Dashboard → SQL Editor:
-- Run: supabase-migrations.sqlThis adds:
skill_reviewstable (1 review per user per skill, star rating + text)version+changelogcolumns on bothskillsandcommunity_skillsupsert_skill_review(),get_skill_reviews(),get_skill_rating()RPC functions
-- In Supabase Dashboard → SQL Editor:
-- Run: supabase-skillforge.sqlThis adds:
skill_compositestable (forged composite skills with provenance + token metrics)skill_graph_edgestable (synergy/overlap edges between skill pairs)mv_popular_skill_pairsmaterialized view (top co-used skill pairs)upsert_graph_edge(),get_top_composites()RPC functions
skill_galaxy/
├── index.html # Main app shell
├── guide.html # Skill creation guide
├── css/
│ └── styles.css # All styles — Claude-inspired theme
├── js/
│ ├── config.js # Supabase client config + environment setup
│ ├── db.js # Skills database (208 hardcoded + 9,872 from Supabase)
│ ├── skills-api.js # Supabase CRUD: community skills, ratings, reviews
│ ├── auth.js # Supabase Auth (login, signup, session)
│ ├── app.js # Frontend logic: render, filter, modals, download, ratings
│ └── icon_renderer.js # Simple Icons + emoji icon rendering
├── api/
│ └── validate-skill.js # Vercel serverless AI validator (Anthropic API)
├── mcp-server/ # MCP server v2.0 for Claude integration
│ ├── index.js # Stdio transport (9 tools, Code Mode, Progressive Disclosure)
│ ├── server-http.js # Streamable HTTP transport (Express, bearer auth, sessions)
│ ├── skills-data.js # Auto-generated skills module (10,171 skills)
│ └── README.md # MCP v2.0 setup instructions
├── supabase-setup.sql # Base database schema
├── supabase-migrations.sql # Ratings, reviews, versioning migration
├── supabase-skillforge.sql # SkillForge: composites + graph edges tables
├── scripts/
│ └── skillforge_benchmark.py # Python benchmark: evolutionary vs. baseline
├── SETUP_GUIDE.md # Full local setup instructions
└── README.md
- Go to skill-galaxy.vercel.app
- Sign in → click Submit a Skill
- Fill in the form or upload a
.mdfile following the Skill Format Guide - Your skill goes live within 24 hours after review
Skill .md format (minimum required frontmatter):
---
name: my-skill-name
description: What this skill does and when Claude should apply it.
version: 1.0.0
---
## Skill Content
Your instructions for Claude go here.Attribution: Your name + profile link appears on every skill card. Contributors retain full ownership — SkillGalaxy is a distribution platform, not a rights holder. You can remove your skill at any time.
- Frontend: Pure HTML, CSS, JavaScript — no frameworks, no build tools
- Backend: Supabase (Postgres + REST API + RLS)
- Hosting: Vercel (static files + serverless API routes)
- Auth: Supabase Auth (email + password)
- AI: Anthropic claude-haiku (optional skill validator)
SkillForge is an evolutionary skill composer that automatically combines multiple skills into optimized composites, with token savings via overlap pruning and synergy-aware selection.
1. Open SkillForge (⚡ button in header)
2. Describe your workflow (e.g., "build ML pipeline with security")
3. SkillForge runs genetic algorithm client-side
4. Returns optimized composite + token savings metrics
| Endpoint | Method | Description |
|---|---|---|
/api/skillforge |
POST | Generate new skill from description |
/api/skillforge-compose |
POST | Compose skills into composite |
/api/forge |
GET | Lattice query with token budget |
- Relevance: TF-IDF cosine similarity to query
- Synergy: Graph edges from shared tags + category
- Fitness:
α·relevance + β·(1-token_penalty) + γ·synergy - Pruning: Overlap removal via desc similarity
Avg Tokens: 4,820 → 1,340 (72% savings)
Accuracy: +11% via synergy-aware selection
Apache 2.0 — free to use, fork, and modify with attribution.
All skill files are free to use, share, and adapt.
timps-website.vercel.app · Product Hunt
If SkillGalaxy is useful to you, consider ⭐ starring the repo and upvoting on Product Hunt — it genuinely helps.