AI Team Orchestration + Quality Gate for Solo Founders
"Build like a team, even when you're alone." — Orchestrate Claude Code sessions as AI agents, decompose goals into tasks, and verify every output with Crewdeck Quality Gate.
Status (2026-07): 개인 운영 도구로 유지·관리된다 (대외 배포/npm publish 계획 없음). 방향 결정의 배경은
docs/design/r3-product-direction.md참고.
npx crewdeckOpens http://127.0.0.1:7200 with a dashboard to manage your AI team.
Crewdeck turns your Claude Code CLI sessions into a team of specialized AI agents.
9 role presets ship out of the box (see templates/agents/):
| Agent | Role |
|---|---|
| CTO | Goal → task decomposition, architecture decisions |
| PM | Planning, prioritization, spec writing |
| Backend / Frontend | Implements features, writes production-ready code |
| UX | UI/UX design, wireframes, terminology |
| QA Engineer | Test strategy, edge cases, regression runs |
| Reviewer | Adversarial review, runs Quality Gate |
| DevOps | Build, CI, release plumbing |
| Marketer | Landing pages, blog posts, SEO content |
Custom agents defined in your project's .claude/agents/*.md take priority over presets.
Every output is independently verified using Crewdeck's Generator-Evaluator separation:
- Generator (Coder) implements the task
- Evaluator (Reviewer) verifies independently — no shared context
- 5-Dimension Verification: Functionality, Data Flow, Design Alignment, Craft, Edge Cases
- Results: PASS / CONDITIONAL / FAIL with severity classification
npxone-line install — SQLite embedded, zero config- Kanban board with drag-and-drop task management
- Project import — analyze local directories, auto-detect tech stack, suggest agents
- GitHub connect — clone repos, auto-analyze, branch strategy
- Goal decomposition — describe what you want, AI breaks it into tasks
- Real-time WebSocket streaming of agent output
- Dark/Light mode with system detection
- Korean/English i18n support
- Command palette (Cmd+K) for quick actions
- Built on Claude Code CLI — uses your existing Claude Pro/Team subscription ($0 extra)
| Paperclip | Crewdeck | |
|---|---|---|
| Quality Gate | None | Generator-Evaluator, 5-dimension |
| Setup | Postgres + onboarding | npx one-line (SQLite) |
| Agent Runtime | Any (Claude, Codex, HTTP) | Claude Code native |
| UX | Functional dashboard | Notion-style, Kanban, dark mode |
| Target | "Autonomous company" (20+) | Solo founders (3-7 agents) |
| Cost | API keys | Claude Pro subscription ($0 extra) |
# Install dependencies
npm install
cd dashboard && npm install && cd ..
# Run dev servers (server + dashboard)
npm run dev
# Type check
npm run typecheck
# Build for production
npm run build
# Start production server
node dist/bin/crewdeck.jscrewdeck/
├── bin/ # CLI entry point (npx crewdeck)
├── server/ # Node.js backend
│ ├── api/ # REST routes + WebSocket
│ ├── core/
│ │ ├── agent/ # Claude Code CLI adapter + session management
│ │ ├── orchestration/ # Goal → Task decomposition + execution
│ │ ├── project/ # Import, GitHub connect, tech stack analyzer
│ │ └── quality-gate/ # Crewdeck 5-dimension verification engine
│ └── db/ # SQLite schema (7 tables)
├── dashboard/ # React + TailwindCSS + Zustand
│ └── src/
│ ├── components/ # 40+ React components
│ └── i18n/ # Korean + English translations
├── shared/ # TypeScript type definitions
└── templates/ # Agent role presets (YAML)
| Layer | Tech |
|---|---|
| Frontend | React, TailwindCSS v4, Zustand, @dnd-kit |
| Backend | Node.js, Express, TypeScript |
| Database | SQLite (better-sqlite3) |
| Real-time | WebSocket (ws) |
| AI Runtime | Claude Code CLI (subprocess, Paperclip pattern) |
| i18n | react-i18next (ko/en) |
| Build | tsup + Vite |
MIT
Inspired by Paperclip (MIT License). Quality Gate 방법론은 Crewdeck 프로젝트에서 출발했으며, 현재는 독립적으로 유지된다.