DevTeam CLI

Create AI development teams that build software together

DevTeam CLI sets up a multi-agent AI development team on ClawdBot. Four specialized AI agents collaborate to plan, build, and security-review your projects—all coordinated through Telegram.

Features

• 4-Agent Team: Planner, Backend, Fullstack, and Security Reviewer
• Orchestrated Workflow: Structured phases with quality gates
• Security First: Dedicated security agent reviews all code
• Multi-Model Support: Different AI models for different tasks
• Real-Time Collaboration: Watch your team work in Telegram
• Human Oversight: Stay in control, approve decisions

Quick Start

# Install
npm install -g devteam-cli

# Initialize a project
devteam init --name my-app

# Configure Telegram
devteam setup telegram

# Start building
devteam start --idea "Build a todo app with user authentication"

Installation

Prerequisites

• Node.js 18+
• ClawdBot installed
• Telegram account
• API keys for AI providers (Anthropic, Google)

Install

npm install -g devteam-cli

Commands

devteam init

Initialize a new dev team project.

devteam init --name my-project
devteam init --name my-project --channel telegram

Creates:

• Agent workspaces with SOUL.md files
• ClawdBot configuration
• Shared workspace for collaboration

devteam setup telegram

Interactive wizard to configure Telegram.

devteam setup telegram
devteam setup telegram --group-id -1001234567890

devteam start

Start the dev team with a project idea.

devteam start --idea "Build a REST API for a blog"
devteam start --idea-file ./project-brief.md
devteam start --idea "..." --dry-run  # Validate without starting

devteam stop

Stop the running dev team.

devteam stop

devteam status

Check project and agent status.

devteam status
devteam status --verbose
devteam status --workflow   # Show phase progress
devteam status --git        # Show git status
devteam status --security   # Show security review

devteam workflow

Manage the development workflow.

devteam workflow report     # Generate status report
devteam workflow advance    # Move to next phase

devteam security

Security review management.

devteam security review     # Run security review
devteam security report     # Show latest report

devteam config

Manage project configuration.

devteam config get                    # Show all config
devteam config get models.backend     # Get specific value
devteam config set models.backend "anthropic/claude-sonnet-4-5"

The Team

Planner (Lead Architect)

• Breaks down requirements into phases
• Assigns tasks to team members
• Tracks progress and handles blockers
• Ensures security review before release

Model: anthropic/claude-opus-4-5 (best reasoning)

Backend Developer

• Designs database schemas
• Builds API endpoints
• Implements authentication
• Writes server-side logic

Model: anthropic/claude-sonnet-4-5 (speed + capability)

Fullstack Developer

• Builds user interfaces
• Connects frontend to backend
• Handles responsive design
• Ensures good UX

Model: google/gemini-2.5-pro (strong at UI)

Security Reviewer

• Reviews all code for vulnerabilities
• Audits dependencies
• Blocks releases with critical issues
• Provides remediation guidance

Model: anthropic/claude-opus-4-5 (deep analysis)

Workflow

DevTeam follows a structured methodology:

Project
├── Milestone 1 (Phases 1-3)
│   ├── Phase 1 → Implement → Analyze → Test → Fix
│   ├── Phase 2 → Implement → Analyze → Test → Fix
│   ├── Phase 3 → Implement → Analyze → Test → Fix
│   └── Milestone Review (Security)
├── Milestone 2 (Phases 4-6)
│   └── ...
└── Final Review

Each phase must pass before advancing. Security reviews happen at milestone boundaries.

Configuration

Project Config (devteam.config.yaml)

project:
  name: "my-project"
  workspace: "/path/to/project"

channel:
  type: "telegram"
  groupId: "-1001234567890"

models:
  planner: "anthropic/claude-opus-4-5"
  backend: "anthropic/claude-sonnet-4-5"
  fullstack: "google/gemini-2.5-pro"
  security: "anthropic/claude-opus-4-5"

security:
  shellSandbox: true
  maxSessionTurns: 50
  auditLog: true

Customizing Agent Roles

Edit the SOUL.md files in agents/*/SOUL.md to customize agent behavior.

Security

DevTeam is designed with security as a priority:

• Input Sanitization: All user input is sanitized before agents see it
• Shell Sandbox: Commands are restricted to a safe allowlist
• Network Egress: Outbound requests limited to approved domains
• Audit Logging: All agent actions are logged
• Security Agent: Dedicated reviewer catches vulnerabilities

Security Checklist

Run before deploying:

devteam security review

Critical checks:

• Gateway bound to loopback
• Auth required for gateway
• Shell sandbox enabled
• Input sanitization active
• Secrets in env vars, not config

Project Structure

my-project/
├── devteam.config.yaml      # Project configuration
├── README.md
├── agents/
│   ├── planner/
│   │   ├── SOUL.md          # Agent personality/instructions
│   │   └── MEMORY.md        # Agent memory
│   ├── backend/
│   ├── fullstack/
│   └── security/
├── shared/
│   ├── PROJECT_STATE.md     # Current status
│   ├── AGENTS.md            # Shared instructions
│   ├── DECISIONS.md         # Architectural decisions
│   └── API.md               # API documentation
├── output/                   # Generated project code
└── .devteam/
    ├── clawdbot.config.yaml # ClawdBot configuration
    ├── runtime-config.yaml  # Runtime configuration
    └── audit/               # Audit logs

FAQ

How do I add more agents?

Edit devteam.config.yaml and add entries to the agents list. Create corresponding agents/<name>/SOUL.md.

Can I use different AI models?

Yes! Configure models in devteam.config.yaml under the models section. Any model supported by ClawdBot works.

How do I customize agent behavior?

Edit the SOUL.md file for each agent. This is their personality and instruction set.

What if agents conflict?

The Planner agent is the coordinator. If conflicts arise, Planner makes the decision or escalates to you.

How do I monitor progress?

• Watch the Telegram group in real-time
• Run devteam status --verbose
• Check shared/PROJECT_STATE.md

Contributing

Contributions welcome! Please read our contributing guidelines.

License

MIT

Built with ❤️ for developers who want AI teammates, not just AI tools.