Until It Works — AI skill for Claude Code
A team of AI coding agents that keeps working until your task genuinely works, with a pixel-art workshop to watch them in.
How to install Until It Works
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open TallblokeUK/until-it-works and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Until It Works does
A team of AI coding agents that keeps working until your task genuinely works, with a pixel-art workshop to watch them in. Works with Claude, Codex, Mercury, Gemini, OpenRouter and local models.
Alternatives in AI
- Zen MCP Server — The power of Claude Code + Gemini / OpenAI / Grok / OpenRouter / Ollama / Custom Model working as one 11.3k ★
- Pixel Agents — Compressed Reference — VS Code extension with embedded React webview: pixel art office where AI agents (Claude Code terminals) are an 6.5k ★
- Qwen Audio Agent — A realtime voice runtime that keeps Agents talking, working, and present 2.3k ★
README
Until It Works(hop)
**A team of AI coding agents that keeps working on your task until it genuinely works, and a pixel-art workshop where you can watch them do it.**
You describe a task. A planner model writes a plan and a definition of "done". Worker models build it, many passes over, or as a swarm working in parallel. Tests, a reviewer, a panel and a final judge check every change, and the work is only approved when all of them agree. If the agents get stuck, they ask you a plain question and wait.
It works with the AI tools you already have: Claude, ChatGPT (Codex), Mercury, Gemini, models through OpenRouter, and models running on your own computer. Nothing touches your files until you choose to keep the result.

What you need
- macOS or Linux
- Python 3.9 or newer and git. A Mac has both once you run
xcode-select --install. - At least one AI coding tool, signed in. For example Claude Code with a Claude subscription, or Codex with a ChatGPT plan. The setup screen shows which tools work with which teams.
- Google Chrome or another Chromium browser. Recommended, so the workshop opens in its own window. Any browser works.
Quick start
**1. Install.**
git clone https://github.com/TallblokeUK/until-it-works.git
cd until-it-works
./install.sh
You should see `installed: mp-agent, mp-status and mp-viz in ~/.local/bin`. If it also says `~/.local/bin is not on your PATH`, add the line it shows to your shell profile and open a new terminal. (`mp-agent` is the command; the name is short for multi-pass agent.)
**2. Set up.**
mp-viz
The workshop opens at http://127.0.0.1:7788 with **SETUP** showing:
- Tools: which AI tools are installed and signed in, and the exact command to fix any that aren't. Press TEST to check one really answers.
- API keys: add any you want to use. They are kept i
Related Skills
Agents Monitor
🏢 Your AI coding agents living in a pixel-art office — watch Claude Code, Codex, Cursor, Gemini & OpenCode wo
VirtualOF
A pixel-art virtual office to command AI coding agents (Claude, Codex) and watch Antigravity.
ZeroCli
🌸 Zero Code: an agentic coding CLI in your terminal, with Zero-chan, a pixel-art anime mascot. It works with
Agent Desktop Pets
Pixel-art desktop pets for Windows - one per AI agent (Claude Code, Codex, Cursor, Copilot, Gemini, Grok, Herm
Pixel Crew
Smart MOD for Claude Code. Picks the cheapest model (Haiku, Sonnet or Opus) for every message and subagent to
AgentDesk
A pixel-art office for your AI coding agents — see every Claude Code / Codex / Gemini CLI session, answer thei
Related Agents
LLM Orchestrator
Use this agent for LLM integration work — prompt engineering, multi-provider abstraction (OpenRouter, Gemini,
Iddy
Iddy, Mellonhead's instructional-designer agent. Builds structured-workshop deliverables (facilitation guide,
Integration Runner
End-of-build closer. Runs the project's real test/validate pipeline and iterates until it is genuinely green,