This project provides a comprehensive AI software team built on Claude Code, addressing the limitations of single AI assistants by orchestrating 119 specialized agents. It targets developers and teams seeking an automated, self-organizing system for complex software development tasks, offering benefits like reduced manual oversight and continuous learning without custom model development.

How It Works

The vibecosystem leverages a swarm of 119 specialized agents, 206 reusable skills, 49 TypeScript hooks (acting as sensors), and 21 behavioral rules. Coordination is achieved implicitly through context injection rather than direct communication. A core innovation is the self-learning pipeline, which automatically converts agent errors into new rules, enhancing the system's knowledge base. "Canavar Cross-Training" ensures that mistakes made by one agent are learned from by the entire team, promoting robust, adaptive AI development. Hooks observe tool calls and inject relevant context, guiding agent actions across five distinct phases: scout, architect, plan, develop, review, test, and learn.

Quick Start & Requirements

• Primary install / run command:

  git clone https://github.com/vibeeval/vibecosystem.git
  cd vibecosystem
  ./install.sh
  

• Non-default prerequisites and dependencies: Claude Code (specifically Claude Max, Opus 4.6 / Sonnet 4.6 models), PostgreSQL + pgvector (Docker).
• Links: Repository itself serves as primary documentation.

Highlighted Details

• Agent Swarm: Over 100 specialized agents coordinate across 5 development phases (Discovery, Development, Review, QA Loop, Final).
• Self-Learning Pipeline: Errors are automatically converted into rules, with patterns promoted to global knowledge for cross-project benefit.
• Dev-QA Loop: Integrated quality gates with automated retries and escalation mechanisms ensure code quality.
• Cross-Project Learning: Knowledge and patterns learned in one project are automatically shared and applied across all projects.
• Canavar Cross-Training: Team-wide error prevention through a mechanism where all agents learn from individual mistakes.

Maintenance & Community

The project is developed by @vibeeval. No other specific community channels, sponsorships, or detailed roadmap information are provided in the README.

Licensing & Compatibility

• License type: MIT License.
• Compatibility notes: The MIT license generally permits commercial use and integration with closed-source projects without significant restrictions.

Limitations & Caveats

The system is entirely dependent on the Claude Code runtime and specific models (Opus/Sonnet), offering no support for custom models or APIs. Its effectiveness is intrinsically linked to the quality of the underlying Claude Code capabilities and the prompt engineering embedded within the agents and skills.