ModelBound Dev Packs

Open, community-curated AI context for software engineering. A collection of production-grade prompts, rules, skills, and memory files that make Cursor, Claude, Copilot, and Windsurf code like your best engineer — every time.

📚 What is this repo?

This is the official open-source home for ModelBound Dev Packs — bundled, version-controlled AI context that teams use to standardize how their AI coding tools behave.

Each folder in this repo is a self-contained Dev Pack containing some combination of:

• System prompts — the role and constraints the AI runs under
• Rules — framework, style, and architecture conventions (Cursor, Copilot, Windsurf formats)
• Skills — reusable, task-specific instructions (refactor, review, write tests…)
• Memory — long-lived project knowledge and decisions
• Evals — test cases to verify the pack actually improves output

The packs in this repo are mirrored to the ModelBound Marketplace, where they can be cloned into any workspace with one click.

📦 Available Dev Packs

Packs are organized by category. Each folder contains its own README.md describing what it does, what files it ships, and how to apply it.

🖼 Frameworks

🔤 Languages

🏛 Architecture

🗄 Data

✅ Quality

⚙️ Process

🚀 Quick Start

Option 1 — Clone from the ModelBound Marketplace (recommended)

1. Sign up at modelbound.co (free tier available)
2. Open the Marketplace
3. Find a pack and click Clone to my workspace
4. (Optional) Connect your repo and one-click sync the pack files into it

You get versioning, AI review, eval scoring, team sharing, and auto-optimization out of the box.

Option 2 — Use the files directly from this repo

git clone https://github.com/modelbound/dev-packs.git
cp -r dev-packs/frameworks/perfect-react-refactor/. /path/to/your/project/

Then point your IDE at the files (see IDE Setup below).

Option 3 — Pull packs on-demand via MCP (zero install)

If your agent supports the Model Context Protocol (Cursor, Claude Code, Claude Desktop, Windsurf, Codex, or any custom agent), you can skip cloning entirely and let the agent fetch packs from ModelBound at runtime.

// .cursor/mcp.json  (or claude_desktop_config.json)
{
  "mcpServers": {
    "modelbound": {
      "url": "https://mcp.modelbound.co",
      "headers": { "Authorization": "Bearer mb_live_YOUR_KEY" }
    }
  }
}

Then ask your agent:

"Load the senior-code-review pack from ModelBound and review this PR."

The agent will call get_pack, pull the latest version, and apply it — no files in your repo, always up to date. See MCP Server below.

🛠 IDE Setup

Dev Packs ship files that map cleanly to every major AI coding tool.

Cursor

• .cursor/rules/*.mdc and .cursorrules are auto-detected
• AGENTS.md is loaded as agent-level context
• Reference any file with @filename in chat

Claude Code / Claude Desktop

claude --system-prompt system-prompt.md

Or load the pack into a Claude Project as knowledge.

GitHub Copilot

• Place copilot-instructions.md at .github/copilot-instructions.md
• Enable custom instructions in VS Code settings

Windsurf

• .windsurf/rules/*.md are auto-loaded
• AGENTS.md provides directory-scoped context to Cascade

Kiro

• .kiro/steering/*.md and .kiro/specs/<feature>/{requirements,design,tasks}.md are auto-loaded

File reference (common across packs)

🔁 How this repo connects to ModelBound

This isn't just a dump of static files — it's the public mirror of a living catalog.

   ┌─────────────────────────┐         ┌──────────────────────────┐
   │ ModelBound Marketplace  │ ◀─────▶ │   github.com/modelbound  │
   │  (clone, version, eval) │   sync  │       /dev-packs         │
   └─────────────────────────┘         └──────────────────────────┘
              ▲                                       ▲
              │ clone / publish                       │ PRs / issues
              ▼                                       ▼
       Your ModelBound workspace          The open-source community
              │
              │ one-click sync
              ▼
        Your Git repo + IDE

• Publish: any pack you build in ModelBound can be pushed to a GitHub repo
• Mirror: the official seed packs in this repo are published & maintained by ModelBound
• Round-trip: approved community PRs here flow back into the Marketplace so every cloner gets the improvement

🔌 MCP Server — use packs without cloning

ModelBound runs a hosted Model Context Protocol server at https://mcp.modelbound.co that exposes every dev pack (and your private library) as live tools your agent can call.

Why use it?

Supported agents

• Cursor & Windsurf — native MCP via .cursor/mcp.json
• Claude Code & Claude Desktop — direct HTTP or mcp-remote proxy
• Codex and any agent harness that speaks JSON-RPC over HTTP
• Custom agents — full tool list at mcp.modelbound.co/docs

Tools relevant to dev packs

• list_packs / get_pack — browse and load any published pack
• search_all — semantic search across packs, rules, skills, and corpora
• list_skills / get_skill — pull individual skills (e.g. "write-pytest-tests")
• add_file_to_pack / create_pack — contribute back from inside your agent
• optimize_tokens / estimate_conversation_cost — keep context lean

Getting an API key

1. Sign up at modelbound.co
2. Go to Settings → API Keys
3. Generate a key with Read scope (or more if you want write access)
4. Drop it into your MCP client config

💡 Read-only keys are safe to share with teammates — they can consume packs but not modify your library.

Full setup guide: [Serve Context to Any AI Agent via MCP →](https://modelbound.co/guides/serve-mcp-context)

🤝 Contributing

We want community contributions. If you have a better React refactor prompt, a smarter SQL reviewer, or a brand-new pack for your favorite stack — open a PR.

Contribute to an existing pack

1. Fork this repo
2. Edit files inside the pack folder (e.g. frameworks/perfect-react-refactor/system-prompt.md)
3. If your change is non-trivial, add or update an eval case in evals/
4. Open a PR with:

• What you changed
• Why it improves the pack (example before/after, eval results, etc.)
• The pack name in the PR title — e.g. [perfect-react-refactor] tighten hook rules

Propose a new Dev Pack

1. Create a new folder under the appropriate category, kebab-case (e.g. frameworks/rust-axum-patterns/)
2. Include at minimum:

• README.md — what it does, who it's for
• system-prompt.md — the core prompt
• One or more rule/skill files
• 2–3 eval cases under evals/

1. Open a PR tagged new-pack

Review & merge flow

1. A maintainer reviews the PR (correctness, scope, safety, style)
2. ModelBound runs the pack's evals against the change
3. On merge:

• The change lands here on main
• The Marketplace pack version bumps automatically
• Existing cloners see an "Update available" notification in their workspace

What we look for

• ✅ Concrete, testable instructions (not vague aspirations)
• ✅ IDE-agnostic where possible; IDE-specific files clearly named
• ✅ No secrets, no proprietary code, no PII
• ✅ Plays well with the rest of the pack (no contradictions)

Code of Conduct

Be kind. Assume good faith. We follow the Contributor Covenant.

🐛 Issues & Feature Requests

• Bug in a pack? Open an issue with the pack name and a reproduction
• Want a new pack? Open an issue with the pack-request label and describe the use case
• Security concern? Email [email protected] directly — please don't open a public issue

🆘 Support & Contact

• 📧 Email: [email protected]
• 🌐 Website: modelbound.co
• 📚 Docs & guides: modelbound.co/guides
• 💬 Community Q&A: modelbound.co/community
• 🛒 Marketplace: modelbound.co/marketplace

🔒 Privacy & Security

• Packs in this repo are 100% open-source and contain no customer data
• When you sync a pack via ModelBound, files only land in repos you explicitly authorize
• ModelBound stores credentials encrypted and only with the scopes you grant
• Full details: Privacy Policy · Terms

📄 License

All packs in this repository are released under the MIT License unless otherwise stated inside an individual pack folder. Use them, fork them, ship them in your products — just don't claim you wrote the originals.

<p align="center"> <b>Built and maintained by <a href="https://modelbound.co">ModelBound</a></b><br> _{The context engineering platform for AI-powered development teams.}<br> <br> <a href="https://modelbound.co">modelbound.co</a> · <a href="mailto:[email protected]">[email protected]</a> </p>