An autonomous multi-agent development pipeline built on OpenClaw.
One prompt in. Working software out. Plan → Build → Review → Test → Deploy — fully autonomous.
You: "Build me a URL shortener with click analytics"
│
┌───────────▼───────────┐
│ 🧠 Architect Agent │ Plans stack, schema, endpoints
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ 💻 Coder Agent │ Implements full project (11 files)
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ 🔍 Reviewer Agent │ Security audit, code quality gate
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ 🧪 Tester Agent │ Writes & runs test suite
└───────────┬───────────┘
│
┌───────────▼───────────┐
│ 🚀 Deployer Agent │ Git init, GitHub push, deploy
└───────────┬───────────┘
│
✅ Live URL + Repo
ClawForge is a skill for OpenClaw that orchestrates five specialized agents to take a project from idea to production:
| Agent | Role | Output |
|---|---|---|
| 🧠 Architect | Stack selection, schema design, API planning | clawforge-state.json with full plan |
| 💻 Coder | Full implementation in TypeScript/whatever fits | Production-ready source files |
| 🔍 Reviewer | Security audit, code quality, best practices | Pass/fail with specific issues |
| 🧪 Tester | Write and run test suite | Test results + coverage report |
| 🚀 Deployer | Git setup, GitHub push, live deployment | Repo URL + live URL |
All agents communicate through a shared state file (clawforge-state.json). If review or tests fail, the pipeline loops back to the Coder with specific fix instructions — up to 2 retries per stage.
From any OpenClaw chat (Telegram, Discord, CLI):
Build me a URL shortener with click analytics
Or explicitly:
ClawForge: Create a habit tracking API with user auth
That's it. ClawForge handles the rest.
clawforge/
├── clawforge.sh # State manager (shared by all agents)
├── clawforge-state.json # Runtime state (agent communication)
├── skills/clawforge/
│ ├── SKILL.md # Orchestrator instructions
│ └── scripts/
│ └── plan.sh # Stage 1: Architect entry point
├── agents/
│ ├── architect/IDENTITY.md # 🧠 Plans the project
│ ├── coder/IDENTITY.md # 💻 Builds the project
│ ├── reviewer/IDENTITY.md # 🔍 Audits the code
│ ├── tester/IDENTITY.md # 🧪 Tests the project
│ └── deployer/IDENTITY.md # 🚀 Ships the project
└── projects/
└── url-shortener/ # Example: built by ClawForge
initialized → architect → coder → reviewer → tester → deployer → complete
↑ ↑ ↑
└──────────┴────────┘ (retry on failure, max 2x)
Each stage updates clawforge-state.json. The orchestrator reads the state to determine what happens next.
# Initialize state for a new project
bash clawforge.sh init "Build a REST API"
# Check current pipeline status
bash clawforge.sh status
# Update any field in state
bash clawforge.sh update .plan.stack.language "TypeScript"
# Record results from each stage
bash clawforge.sh stage architect plan_complete
bash clawforge.sh review-result true ""
bash clawforge.sh test-result 10 8 2
bash clawforge.sh deploy-result https://live-url.com https://github.com/...The first project built by ClawForge — a full URL shortener with click analytics:
- Stack: TypeScript, Express.js, SQLite, nanoid
- Files: 11 source files, fully typed
- Tests: 10 tests, all passing
- Features: Short URLs, custom codes, click tracking, analytics dashboard
→ Live Demo
→ Source Code
→ Audit Report (21 issues found and documented)
- OpenClaw — The AI agent platform
- 5 specialized agents — Each with a focused identity and checklist
- 1 Telegram message — The only human input
MIT