Skip to content

QUICKSTART.md

Latest commit

 

History

History
211 lines (147 loc) · 5.29 KB

QUICKSTART.md

File metadata and controls

211 lines (147 loc) · 5.29 KB

Quick Start — AI Dev Team

Environment

  • Linux (Ubuntu, Debian, or any Linux with bash)
  • Claude Code CLI with Pro or Max subscription
  • Python 3.10+

Recommended: Run AI Dev Team in an isolated environment — a dedicated VM, LXC container, or Docker container. The agents have access to shell tools (Bash, Edit, Write) and will create files, run commands, and modify your workspace. An isolated environment ensures they cannot accidentally affect your host system or other projects.

1. Install

Claude Code CLI

AI Dev Team runs on Claude Code — Anthropic's CLI for Claude AI. You need a Pro ($20/mo) or Max ($100/mo) subscription.

# Install Claude Code
npm install -g @anthropic-ai/claude-code

# Login (opens browser)
claude login

# Verify
claude --version

Python Dependencies

# Clone the project
git clone https://github.com/CelaenoIndustry/OpenSepia.git
cd ai-team

# Install dependencies
pip install -r requirements.txt

Note: AI Dev Team does NOT use the Anthropic API. It calls claude --print directly, which uses your Claude subscription. No API key needed.

2. Set Up Your Project

Option A: Use Claude Code (recommended)

The easiest way to get started — just open Claude Code in the project directory and tell it what you want to build:

cd ai-team
claude

Then tell Claude what to do:

> Initialize an AI dev team project for a REST API with FastAPI and PostgreSQL.
  Set up GitLab integration with token glpat-xxx on gitlab.example.com,
  project group/my-api. Start the team and set up cron.

Claude Code reads the CLAUDE.md instructions and knows how to:

  • Initialize the project (init_project.py)
  • Configure GitLab/GitHub integration (.env + init_integrations.py)
  • Write a detailed brief to the PO agent's inbox
  • Start the first cycle
  • Set up cron for automated runs

This is how the RAG example was built — a single conversation with Claude Code set up the entire project and team.

Option B: Manual setup

# Initialize project
python scripts/init_project.py "My API" \
  "REST API for document management with FastAPI backend and PostgreSQL"

# (Optional) Write a detailed brief to the PO
cp examples/rag-app/po-brief.md board/inbox/po.md
# Edit board/inbox/po.md with your own project brief

# (Optional) Set up GitLab/GitHub integration
cp config/.env.example config/.env
nano config/.env  # fill in your tokens
python scripts/init_integrations.py

Tip: See examples/rag-app/po-brief.md for a real PO brief that produced 3,600 lines of code overnight.

3. Run Your First Cycle

# Core dev team (6 agents): PO, PM, Dev1, Dev2, DevOps, Tester
./scripts/orchestrator_cli.sh dev-team

# Minimal team (3 agents): PO, Dev1, Tester — saves rate limits
./scripts/orchestrator_cli.sh minimal

# Full team including security (9 agents)
./scripts/orchestrator_cli.sh all

# Security audit only (3 agents)
./scripts/orchestrator_cli.sh security

# Single agent (for debugging)
./scripts/orchestrator_cli.sh dev1

4. Set Up Automated Runs

# Edit crontab paths first
nano ai-team.crontab

# Install crontab
crontab ai-team.crontab

Or add manually:

crontab -e

# Dev team every 40 minutes
*/40 * * * * /path/to/ai-team/scripts/orchestrator_cli.sh dev-team >> /path/to/ai-team/logs/cron.log 2>&1

# Security audit once daily at 6:00 AM
0 6 * * * /path/to/ai-team/scripts/orchestrator_cli.sh security >> /path/to/ai-team/logs/cron.log 2>&1

Now go to sleep. The team works while you're away.

5. Check Results

# Sprint board
cat board/sprint.md

# Backlog
cat board/backlog.md

# Generated code
ls workspace/src/

# Standup reports
cat board/standup.md

# Agent messages
cat board/inbox/*.md

# Live monitoring
tail -f logs/cron.log

GitLab / GitHub Integration

# Configure tokens
cp config/.env.example config/.env
nano config/.env  # fill in your tokens

# Initialize integration (creates labels, boards, clones repo)
python scripts/init_integrations.py

With integration enabled you get:

  • Issues on a Kanban board
  • Merge Requests / Pull Requests from developers
  • Agent comments on issues
  • Code review comments posted directly on MRs
  • Automatic MR approval when agents approve code
  • Auto-merge of approved MRs/PRs

Human Intervention

Send messages to any agent by writing to their inbox:

echo "## Message from Human
STORY-003 priority is now CRITICAL!" >> board/inbox/pm.md

The agent reads your message in the next cycle. You can also comment directly on GitLab/GitHub issues — agents read those too.

Rate Limits

Plan Messages/5h Recommended cycles/day Team size
Pro ($20/mo) ~45 3-4 minimal (3)
Max ($100/mo) ~225 15-20 dev-team (6) or all (9)

Troubleshooting

"Rate limit reached"

Wait for the rate limit to reset (~5 hours) or reduce team size to minimal.

Claude Code not logged in

claude login
# Follow browser instructions

Agents produce poor output

  • Check board/project.md — is the project well described?
  • Check config/project.yaml — is the tech stack correct?
  • Try running a single agent first: ./scripts/orchestrator_cli.sh po
⚡