Back to directory

metaswarm

A multi-agent orchestration framework for Claude Code, Gemini CLI, and Codex CLI that coordinates specialized agent personas through a structured software development lifecycle, from research to a merged pull request, with mandatory review gates and test-driven development.

metaswarm

metaswarm coordinates specialized agent personas, such as researcher, architect, coder, security auditor, and PR shepherd, and a set of orchestration skills through a nine-phase workflow: research, plan, a design review gate, work-unit decomposition, orchestrated execution, final review, PR creation, PR shepherding, and closure with learning capture. Each work unit runs an implement, validate, adversarial-review, commit loop, with the orchestrator independently validating results rather than trusting subagent self-reports.

A parallel design review gate runs several specialist agents in parallel with a capped number of iterations before escalating to a human. Recursive orchestration lets swarm coordinators spawn issue orchestrators and sub-orchestrators for complex epics, and a git-native knowledge base of facts, gotchas, decisions, and anti-patterns is primed before every task. It can optionally delegate implementation or review to other CLI agents for cost savings and cross-model review, and can capture screenshots for reviewing web UIs.

It is written in Shell and MIT licensed.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.