AI Agent Orchestration System
Production-grade multi-agent framework for automated issue detection and resolution using Claude Code.
Run 40+ AI agents in parallel — each specialized by domain — coordinated by an orchestrator that uses only 3,000 tokens instead of 265,000.
Security Notice
Do not run this framework in automated CI/CD with access to production secrets.
This framework lets AI agents modify code autonomously. Permission requests timeout without user response, which defeats safety gates in unattended environments. Run locally with human oversight.
Before running on any codebase:
- Review all issue files before
/fix-all (issue files are prompt-injection vectors)
- Run without production credentials (strip
.env first if running on a real project)
- Use branch protection on main; don't let agents push directly
- Validate issue files with
python3 tools/validate_issue_file.py issues/ before running fixers
See SECURITY.md for the full threat model.
Requirements
- Claude Code CLI installed (this framework uses Claude Code slash commands and Task tool)
- Claude Sonnet (orchestrator agents — Haiku cannot use Task tool)
- Claude Haiku (lane agents — cost-optimized for parallel execution)
- Python 3.9+
- macOS or Linux (Windows not supported — shell scripts assume POSIX)
- Git (framework uses git hooks and history)
Quick Start
# Clone and enter the repo
git clone https://github.com/jonathanmejia4/ai-agent-orchestration-system.git
cd ai-agent-orchestration-system
# Install dependencies
pip install -r requirements.txt
# Hunt for issues across all lanes
/find-all
# Fix all discovered issues
/fix-all
# Verify fixes are real
/verify-fixes
What This Does
See it work: examples/QUICK_START_WALKTHROUGH.md — real terminal output from a live run.