metaswarm  by dsifry

A self-improving multi-agent framework for autonomous software development

Created 3 months ago
264 stars

Top 96.5% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

Metaswarm is a production-tested, self-improving multi-agent orchestration framework automating the software development lifecycle. It empowers engineers by coordinating specialized AI agents for tasks from research to PR management, enforcing rigorous quality gates and learning from each session to deliver production-level code with high test coverage.

How It Works

The system orchestrates 18 specialized AI agents and 13 skills through a 9-phase workflow, featuring a 4-phase "Orchestrated Execution Loop" (Implement, Validate, Adversarial Review, Commit). Core principles include recursive orchestration, parallel design review gates, and a git-native knowledge base managed via BEADS CLI for context priming. Optional delegation to external AI tools like Codex CLI and Gemini CLI enables cross-model adversarial review, enhancing robustness.

Quick Start & Requirements

Installation varies by AI CLI:

  • Claude Code: claude plugin install metaswarm then /setup.
  • Gemini CLI: gemini extensions install https://github.com/dsifry/metaswarm.git then /metaswarm:setup.
  • Codex CLI: curl -sSL https://raw.githubusercontent.com/dsifry/metaswarm/main/.codex/install.sh | bash then $setup. Cross-platform: npx metaswarm init. Prerequisites: Node.js 18+, BEADS CLI v0.40+ (recommended), GitHub CLI (recommended). Playwright optional for visual review. See INSTALL.md for details.

Highlighted Details

  • Mandatory Test-Driven Development (TDD) and 100% test coverage enforcement.
  • Self-learning system reflects on feedback, failures, and user interactions to update its knowledge base.
  • "Trust Nothing, Verify Everything" via independent validation and cross-model adversarial reviews.
  • Selective knowledge priming using BEADS CLI optimizes context window usage from a growing knowledge base.

Maintenance & Community

Created by Dave Sifry, extracted from a production SaaS codebase. Builds on BEADS (git-native issue tracking) and Superpowers (agentic skills). No specific community channels listed.

Licensing & Compatibility

MIT license. Permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

Visual review requires optional Playwright installation. BEADS CLI and GitHub CLI are recommended for full task tracking and PR automation. Effectiveness depends on underlying external AI CLIs.

Health Check
Last Commit

1 week ago

Responsiveness

Inactive

Pull Requests (30d)
4
Issues (30d)
5
Star History
58 stars in the last 30 days

Explore Similar Projects

Feedback? Help us improve.