Godmode Lite is a free Claude Code skill. It wraps any task and drives Claude through 4 execution layers instead of a single unstructured attempt: deep context loading, complete execution, exhaustive testing, and polish. A small zero-dependency Node CLI under runner/ gives each layer a deterministic pass/fail gate, so Claude can't skip a step or call something done before it verifiably is.
- Claude Code, OpenAI Codex, Cursor
- Windows, macOS, Linux (Node 18+)
- Run state lives in a per-user folder, never inside the skill folder. In a sandboxed host that only allows writes in the project (Codex), it goes to
.evo/godmode-lite/in the project. - If the host has no question tool, the skill asks in plain text in the chat.
Clone the repo into ~/.claude/skills/godmode-lite/:
git clone https://github.com/Getgodmode/godmode-lite.git ~/.claude/skills/godmode-lite
chmod +x ~/.claude/skills/godmode-lite/runner/bin/liteFor Codex, clone into ~/.codex/skills/godmode-lite/ instead. The skill finds its runner next to SKILL.md, wherever it lives.
No git? Download and extract the tarball instead:
mkdir -p ~/.claude/skills/godmode-lite
curl -sL https://github.com/Getgodmode/godmode-lite/archive/refs/heads/main.tar.gz \
| tar -xz --strip-components=1 -C ~/.claude/skills/godmode-lite
chmod +x ~/.claude/skills/godmode-lite/runner/bin/literunner/bin/lite requires the full runner/src/ tree, so curling individual files won't work — always pull the whole repo.
Type any of these before a task in Claude Code:
godmode lite: build a REST API with auth
gm lite: refactor this module with full test coverage
try godmode lite: audit this codebase and fix every issue
Claude then runs the runner/bin/lite CLI through each layer in order (start → discover → check → test → polish → end), reading the envelope each command returns to decide the next step. See SKILL.md for the full protocol.
| Layer | What it does |
|---|---|
| Deep Context Load | Reads every related file before writing a line |
| Complete Execution | Every edge case handled, no stubs, no TODOs |
| Exhaustive Testing | Detects and runs the project's test suite, fails the gate on a failing test |
| Polish | Formatter/linter pass scoped to files the run touched |
The full Godmode protocol adds security hardening, alternative-approach exploration, auto-documentation, and ripple checking, plus higher tiers with smarter planning, a self-improving evolution engine, and an execute-assess-fix loop that won't stop until every dimension passes.
Need something built to spec instead of a general-purpose skill? See the custom build offer.
MIT — see LICENSE.