Skip to content
All projects

Loopforge

Early — not production ready

A game-development toolkit made of Agent Skills and a deterministic CLI.

Loopforge helps an existing coding agent turn a game idea into tested, evidence-backed playable builds without hiding project state inside a chat. Creative judgment stays in the skills and in human review; state transitions, validation, evidence and recovery stay in the CLI. The project is in early implementation.

Overview

The split is the point. Skills hold creative judgment and human review; the CLI owns state transitions, validation, evidence and recovery. Rather than building a proprietary agent runtime, Loopforge drives coding agents you already run: `loopforge setup --host codex` installs the official skills into the Codex skills directory, and `$loopforge-router` reads durable project state to route the next action to gameplay prototyping, Godot implementation, game design or art production.

Project state is durable and inspectable instead of living in a chat log: hash-chained events, locking, reconciliation, read-only diagnostics, artifact integrity validation, registered evidence, hypothesis records and guarded stage advancement. `status` also derives scoped quality claims and marks them stale when the source identity changes, so a claim cannot outlive the thing it was made about.

The repository is explicit about what is not done. This is early implementation: validation against a real Godot installation and production-stage release skills are not implemented yet. The package is not published to PyPI either — `uv tool install loopforge` does not refer to this project — so install from the Git URL and pin a reviewed tag or commit for reproducible environments.

What it does

  • Agent Skills plus a deterministic Python CLI: creative judgment in the skills, state transitions and validation in the CLI.
  • Hash-chained event state with locking, reconciliation and recovery, so project history is auditable rather than conversational.
  • Evidence-first workflow: registered evidence, hypothesis records, artifact integrity validation and guarded stage advancement.
  • Scoped quality claims that go stale automatically when the identity of their source changes.
  • Godot 4 build/test adapter, playtest import, and atomic keep, kill or refactor decisions on prototypes.
  • Drives coding agents you already use instead of shipping a proprietary agent runtime.

Get started

Requirements: Python 3.11+, uv and Git; Godot 4 only for the Godot build workflow. Not published to PyPI — install from the Git URL.

Install the CLI

uv tool install git+https://github.com/dopejs/loopforge.git

Install the skills into Codex

loopforge setup --host codex