GodModeSkill

🧠 GodModeSkill
/work — a single Claude Code command that runs every plan, feature, and bugfix past 3 different model families before any merge.

Codex + Gemini + OpenCode (Kimi/DeepSeek) review your work in parallel. Quorum must agree, or Claude revises and retries. Zero token-burn while waiting. Auto-handles git. Always asks before merge.

👉 Looking for the productized version? Use Chorus

GodModeSkill was the bash prototype that proved the multi-LLM peer-review idea. We rebuilt it as a polished cross-platform product:

🎶 Chorus — npm install -g chorus-codes

- Polished cross-platform UI (cockpit at localhost:5050) for fan-out reviews + run inspection
- Works with any AI CLI (Claude Code, Codex, Gemini, OpenCode, Kimi, Cursor, Windsurf) via MCP
- OpenRouter built in, plus per-voice fallback chains, voice auto-disable on quota, and live cost tracking
- Apache-2.0, open source:

GodModeSkill stays here as the original /work bash skill for Claude Code users who prefer a script. New features land on Chorus.

✨ Why this exists

🤖 Single-model AI coding fails silently. Code looks reasonable, tests pass — but a subtle bug or design drift only shows up later.

👯 Same-family models share blind spots. Three Codex sessions reviewing each other is an echo chamber.

✅ Lineage diversity is the unlock. A different model family reading the same code from scratch catches a startling amount of what the author missed.

🚀 Quick start

Option 1 — paste-and-go (recommended)

Open Claude Code and paste INSTALL.md. Claude asks which CLIs you have, then installs everything.

Option 2 — manual

🎯 Modes

Claude picks the mode, picks 3 reviewers, builds the pack, runs the loop, and asks for sign-off at the merge gate.

📋 Prerequisites

You need at least 3 different model families:

- 🟣…