automagik-dev/genie

Wishes in, PRs out. CLI agent that interviews you, plans the work, dispatches parallel agents in isolated worktrees, and reviews code before you see it.

What it solves

Genie provides a structured, agentic workflow for software development, ensuring that AI agents don't just write code, but follow a rigorous process of brainstorming, planning, and independent review. It prevents the "hallucination" of progress by requiring that every gate (review) is performed by an agent that did not perform the work, and keeps all designs and plans as versioned markdown files in the repository itself to maintain long-term context.

How it works

Genie operates as a set of skills and saved workflows that integrate with AI coding agents (like Claude Code or Codex). It uses a specific loop to handle tasks:

  • Brainstorming: Ideas are settled into a DESIGN.md file.
  • Wishing: Single tasks are handled via the /wish command, which builds the change in a separate worktree, runs the repository's own checks, and opens a pull request.
  • Working: Larger tasks are broken into execution groups in a WISH.md file and executed in dependency order by scoped workers.
  • Reviewing: Every stage is gated by an independent reviewer agent that must return a verdict (e.g., SHIP, FIX-FIRST) before the work can proceed.

Task state is tracked in a local SQLite database (.genie/genie.db) and can be viewed as a kanban board via the CLI.

Who it’s for

Developers who use AI coding agents and want a professional software engineering process—including design reviews, independent verification, and git-based context—rather than than just raw code generation.

Highlights

  • Independent Review Gates: Ensures a separate agent performs the review to avoid self-approval.
  • Git-based Context: Designs and plans are stored as markdown in the repo, making them the source of truth for any agent.
  • Worktree Isolation: Tasks are built and tested in their own worktrees to avoid polluting the local environment.
  • Integrated Skill Catalog: Ships with 18 specialized skills (e.g., research, docs-audit, council) and 12 saved workflows.
  • Agent Agnostic: Works across different agent harnesses like Claude Code and Codex.

Related

  • Project
    coollabsio/jean

    Jean is a desktop‑or‑server workspace that lets developers run multiple AI coding agents (Claude, Codex, Grok, etc.) in parallel, each inside its own git worktree. It bundles project/worktree management, session tracking, built‑in terminal, diff viewer, and deep GitHub/Linear/Sentry integrations, with a web UI for remote access.

  • Project
    Owl-Listener/designer-skills

    A set of Claude Code/Gemini‑CLI plugins that turn design knowledge into machine‑readable skills and ready‑to‑run commands, letting AI agents execute full design workflows (research, strategy, UI, interaction, systems, ops, critique) directly from the assistant.

  • Project
    agent-sh/agentsys

    A modular runtime and orchestration system for AI agents that automates the software development lifecycle, from task discovery to PR creation and delivery validation.

  • Project
    EveryInc/compound-engineering-plugin

    Compound Engineering is a cross‑platform plugin that equips AI coding agents with 36 skills to run a structured “brainstorm → plan → work → simplify → review → compound” loop, persisting learned insights for future AI‑driven development.

  • Project
    kaankiziltug/logo-design-skill

    A plug‑in (Agent Skill) that lets Claude, Gemini, Codex, Cursor, etc., act as a full‑featured logo designer. It follows a brief → research → concept → audit → checkpoint → brand‑kit workflow, using a Python toolset and a curated library of 1 400+ SVG logos. Install via Claude’s marketplace or by copying the skill folder, then ask the agent for a logo and receive production‑ready SVGs, favicons, mock‑up boards and a style guide.