"Low-cost tasks to affordable agents; high-level reasoning to top-tier models."
Multi-agent orchestration plugin for OpenCode. Coordinate specialized AI agents working together like a professional development team.
- Introduction
- Core Philosophy
- The Crew
- Features
- Installation
- Quick Start
- Usage
- SOPs
- Configuration
- CLI Commands
- Development
- Testing
- Contributing
- License
crew-opencode is a multi-agent orchestration system that coordinates specialized AI agents to work together on software development tasks. Instead of using a single expensive model for everything, crew-opencode strategically assigns tasks to the right agent with the right model, optimizing both cost and quality.
When you type the crew command, you become the Project Manager, directing a team of specialized agents that follow structured Standard Operating Procedures (SOPs) to deliver high-quality results.
-
Cost-Effective: Simple, repetitive tasks go to lightweight models (Haiku), while complex reasoning uses top-tier models (Opus). This approach reduces costs by 60-70% compared to using premium models for everything.
-
Specialization: Each agent is an expert in their domain (PM, Technical Analysis, Frontend, Design, QA) with tailored prompts and optimal model selection.
-
Accountability: When agents fail, they generate Incident Reports ("Apology Letters") with root cause analysis, risk assessment, and prevention strategiesβensuring continuous improvement.
Five specialized agents working together:
| Agent | Role | Model | Cost | Specialty |
|---|---|---|---|---|
| PM | Project Manager | Claude Opus 4.5 | π°π°π° | Orchestration, strategy, planning |
| TA | Technical Analyst | Claude Sonnet 4.5 | π°π° | Research, architecture, analysis |
| FE | UI/UX Engineer | Gemini 3 Pro | π°π° | Frontend implementation |
| Design | Designer | GPT 5.2 Medium | π°π° | UX flows, design systems |
| QA | Quality Assurance | Claude Haiku 4.5 | π° | Testing, verification |
Cost Optimization: By using Haiku for QA (repetitive testing) and Opus only for PM (critical decisions), you get premium quality at a fraction of the cost.
- 5 specialized agents with role-specific prompts and optimal models
- Parallel execution for tasks that can run simultaneously
- Dependency management ensures tasks complete in the correct order
- Feature Development (60-90min): Full feature implementation with research, design, coding, and testing
- Bug Fix (40-70min): Root cause analysis and minimal, targeted fixes
- Refactoring (100-160min): Safe refactoring with comprehensive test coverage
- Automatic generation when tasks fail
- Root cause analysis: Why did it fail?
- Risk assessment: What's the impact?
- Prevention strategy: How to avoid it next time?
- Custom tools for orchestration, status checking, and reporting
- Hooks for pre/post tool execution and session cleanup
- Seamless installation to OpenCode's plugin system
- Real-time progress tracking with event-driven updates
- Color-coded output for easy reading
- Dry-run mode to preview execution plans
- Comprehensive help and error messages
- Bun >= 1.0 or Node.js >= 18.0 (for npm installation)
- OpenCode (optional, for plugin integration)
- API keys for LLM providers (Anthropic, OpenAI, Google)
# Configure npm to use GitHub Packages
echo "@sehyun0518:registry=https://npm.pkg.github.com" >> ~/.npmrc
# Install globally
npm install -g @sehyun0518/crew-opencodeDownload the latest binary for your platform from GitHub Releases:
macOS (ARM64 - M1/M2/M3)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-macos-arm64 -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/macOS (Intel)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-macos -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/Linux (x64)
curl -L https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-linux-x64 -o crew-opencode
chmod +x crew-opencode
sudo mv crew-opencode /usr/local/bin/Windows (x64)
# Download from: https://github.com/sehyun0518/crew-opencode/releases/download/v1.0.0/crew-opencode-windows-x64.exe
# Add to PATH or run directlybunx @sehyun0518/crew-opencode installAfter installation, register with OpenCode:
# Install globally to ~/.opencode/
crew-opencode install --global
# Or install locally in current project
crew-opencode install --localnpm install -g crew-opencodecrew-opencode crew "Add authentication to the API"crew-opencode listcrew-opencode reportscrew-opencode crew "your task description"# Feature development (default)
crew-opencode crew "Add user profile page"
# Bug fixing
crew-opencode crew "Fix login timeout issue" --sop bugfix
# Refactoring
crew-opencode crew "Refactor authentication service" --sop refactorPreview the execution plan without running:
crew-opencode crew "Add comments feature" --dry-runAdding a New Feature:
crew-opencode crew "Add dark mode toggle to settings"Output:
π― crew-opencode - Multi-Agent Orchestration
Task: Add dark mode toggle to settings
SOP: feature
Mode: Live
π Execution Plan:
Workflow: feature
Steps: 5
Required Agents: PM, TA, DESIGN, FE, QA
1. PM
Analyze requirements and create execution plan
2. TA + DESIGN [Parallel]
β’ TA: Research documentation and analyze codebase
β’ DESIGN: Review UI/UX flows and propose design
3. FE
Implement feature based on specs
4. QA
Write and run tests, verify quality
5. PM
Final review and summary
π Starting execution...
Fixing a Bug:
crew-opencode crew "Fix navbar overflow on mobile" --sop bugfixRefactoring Code:
crew-opencode crew "Extract API client into separate module" --sop refactorWhen to use: Implementing new features
Workflow:
- PM: Requirements analysis and planning
- TA + Design (Parallel): Technical research and UX design
- FE: Implementation
- QA: Testing (β₯80% coverage)
- PM: Final review
Success Criteria:
- All requirements met
- 80%+ test coverage
- No critical bugs
- PM approval
When to use: Fixing bugs or issues
Workflow:
- PM: Bug analysis and prioritization
- TA: Root cause investigation
- FE: Minimal fix implementation
- QA: Regression testing
- PM: Review and closure
Principles:
- Fix root cause, not symptoms
- Minimal changes only
- Always add regression test
When to use: Improving code quality
Workflow:
- PM: Define scope and goals
- TA: Architecture analysis
- QA: Create safety net (baseline tests)
- FE: Incremental refactoring
- QA: Verify no regressions
- PM: Final validation
Safety First:
- Tests before refactoring
- Incremental changes
- Continuous verification
Configuration file: .opencode/crew-opencode/crew-opencode.json
{
"version": "1.0",
"crew": {
"pm": {
"enabled": true,
"model": "claude-opus-4.5",
"maxTurns": 10,
"temperature": 0.7
},
"ta": {
"enabled": true,
"model": "claude-sonnet-4.5",
"maxTurns": 15,
"temperature": 0.5
}
// ... other agents
},
"sop": {
"default": "feature"
},
"incidentReport": {
"enabled": true,
"outputDir": "reports",
"format": "json"
}
}Each agent can be configured:
enabled: Enable/disable the agentmodel: Model to usemaxTurns: Maximum conversation turnstemperature: Model temperature (0-2)
# Execute a task with the crew
crew-opencode crew "task description" [options]
# Install plugin to OpenCode
crew-opencode install [--global|--local]
# Uninstall plugin
crew-opencode uninstall [--global|--local]# List available agents and SOPs
crew-opencode list [--agents|--sops]
# Diagnose installation
crew-opencode doctor
# View incident reports
crew-opencode reports [--limit N]
# Manage configuration
crew-opencode config [key] [value]-s, --sop <type>: SOP type (feature, bugfix, refactor)--dry-run: Preview execution plan-g, --global: Global installation-l, --local: Local installation
# Clone repository
git clone https://github.com/sehyun0518/crew-opencode.git
cd crew-opencode
# Install dependencies
bun install
# Build
bun run build
# Run locally
bun run devcrew-opencode/
βββ src/
β βββ agents/ # Agent definitions (PM, TA, FE, Design, QA)
β βββ cli/ # CLI commands
β βββ config/ # Configuration system
β βββ core/ # Orchestration engine
β βββ hooks/ # OpenCode hooks
β βββ sop/ # SOP workflows
β βββ tools/ # Custom tools
βββ tests/ # Test suite
βββ templates/ # Templates
# Run all tests
bun test
# Run with coverage
bun run test:coverage
# Watch mode
bun test --watchCoverage Goals: β₯80% (lines, functions, statements)
We welcome contributions! See CONTRIBUTING.md for guidelines.
- Fork the repository
- Create a feature branch (
git checkout -b feature/amazing-feature) - Make your changes
- Add tests
- Ensure tests pass (
bun test) - Commit (
git commit -m 'feat: add amazing feature') - Push (
git push origin feature/amazing-feature) - Open a Pull Request
- Follow Conventional Commits
- Maintain β₯80% test coverage
- Update documentation
- Add yourself to contributors list
MIT Β© 2026 crew-opencode
See LICENSE for details.
- Built with Bun and TypeScript
- Powered by Claude, Gemini, and GPT
- Inspired by collaborative software development teams
Made with β€οΈ by the crew-opencode team
- Root Cause: Why did it stop?
- Risk Analysis: What impact does this error have on the project?
- Prevention Strategy: How will we ensure this doesn't happen again?
- This is not just a log; it is a Self-Reflection process for the agent.
OpenCode must be installed first.
# Install OpenCode first
curl -fsSL https://opencode.ai/install | bash
# Install crew-opencode plugin
bunx crew-opencode install# Run with default workflow
opencode
# Or execute a specific workflow
bunx crew-opencode run featureWe are striving to create a better agent-based coding environment. We acknowledge that our current methods may not be the best, and we eagerly await your wisdom.
- Proposals for new Roles
- Improvements in Prompt Engineering
- Bug reports and fixes
- Etc.
All PRs are welcome. Please join us in becoming a constantly growing organization!
Β© 2026 crew-opencode. All rights reserved.
