by Innate-Labs
Local-first desktop agent that turns a prompt into a reviewed, playable browser game with Codex App Server.
# Add to your Claude Code skills
git clone https://github.com/Innate-Labs/Noobi.aiLast scanned: 8/28/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-08-28T15:55:31.180Z",
"npmAuditRan": true,
"pipAuditRan": true,
"promptInjectionRan": true
}See how Noobi.ai compares with popular alternatives.
Noobi.ai is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by Innate-Labs. Local-first desktop agent that turns a prompt into a reviewed, playable browser game with Codex App Server. It has 247 GitHub stars.
Yes. Noobi.ai passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/Innate-Labs/Noobi.ai" and add it to your Claude Code skills directory (see the Installation section above).
Noobi.ai is primarily written in TypeScript. It is open-source under Innate-Labs on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh Noobi.ai against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.

Noobi.ai gives Codex a bounded game-production loop instead of asking one agent to improvise everything in a single pass. A read-only Planner scopes the work, an Implementer builds inside an isolated project directory, an independent Reviewer checks the result, and the host builds and plays the production output before completion. Failed review, build or playtest checks return to the same Implementer for up to three repair rounds.
Current status: developer preview for macOS. No signed or notarized binary is published yet; run it from source. The output is a standalone Web or Godot 4 project in your own workspace.
| What you get | |
|---|---|
| From prompt to playable | Start with a natural-language brief, let the Agent choose Web or Godot 4, and preview a real standalone game workspace locally. |
| Review-gated production loop | Planner → Implementer → Reviewer → formal build → automated playtest. Failed checks trigger up to three bounded repair rounds. |
| Automatic experience evaluation | A sandboxed hidden browser exercises core controls, captures evidence, and checks visible gameplay, animation continuity and runtime errors. |
| Real media pipeline | Route images, music, speech, sound effects and 3D through configured providers, Codex ImageGen, imported assets or procedural fallbacks. Failed assets remain visible as retryable placeholders. |
| A workspace you own | Every game is a normal local project. Inspect the files, continue with Codex, commit it to Git, or take it outside Noobi.ai. |
| Built to extend | Add media providers, Codex Skills, MCP servers, agent prompts, project templates, Godot export targets, or an entirely different workbench UI. |
git clone https://github.com/Innate-Labs/Noobi.ai.git
cd Noobi.ai
npm ci
npm run dev
On first launch, open Settings → Codex account and sign in. Noobi.ai uses an app-private userData/codex-home; it does not overwrite your global ~/.codex configuration.
If Codex cannot be located automatically, point Noobi.ai at a binary explicitly:
NOOBI_CODEX_BIN=/absolute/path/to/codex npm run dev
flowchart LR
Idea["Idea + reference files"] --> Preflight["Capability and engine preflight"]
Preflight --> Plan["Planner<br/>read-only"]
Plan --> Build["Implementer<br/>workspace-write"]
Build --> Review["Reviewer<br/>read-only"]
Review --> Pass{"Pass?"}
Pass -- "No" --> Repair["Repair<br/>up to 3 rounds"]
Repair --> Review
Pass -- "Yes" --> EngineBuild["Web build or<br/>Godot import + export"]
EngineBuild --> Playtest["Hidden-browser playtest<br/>inputs + screenshots + errors"]
Playtest --> Gate["Host proof gate"]
Gate -. "Repairable finding" .-> Repair
Gate --> Done["Playable local project"]
Gate -. "External blocker" .-> Blocked["Blocked"]
The workbench visualizes Brief → Scaffold → GDD → Assets → World → Code → Verify → Complete. Those stages explain progress; the Reviewer and host proof gate decide whether a run is actually complete.

Noobi.ai is intentionally organized around replaceable boundaries. A useful fork can start small:
| Goal | Start here |
|---|---|
| Add or change a media provider | mediaProviderStore.ts and mediaGenerationService.ts |
| Connect a tool or internal service through MCP | mcpConfigManager.ts |
| Change Planner, Implementer, Reviewer or Repair behavior | promptTemplateStore.ts and the prompt contracts in gameHarness.ts |
| Change the generated game scaffold and project rules | workspaceTemplate.ts |
| Build a new production experience | src/renderer/components |
| Add a new host-side dynamic tool | mediaToolBroker.ts |
Good first directions include a new provider adapter, Windows/Linux packaging, sample-game galleries, accessibility improvements, and additional deterministic game templates. See the roadmap and contribution guide.
safeStorage and never returned to the Renderer after saving127.0.0.1Read the full product capability map and architecture guide.
flowchart LR
User["Creator<br/>prompt + assets"] --> Renderer["React Renderer<br/>workbench · approvals · preview"]
subgraph Desktop["Electron desktop"]
Renderer <-->|"typed IPC"| Main["Electron Main<br/>trusted host"]
Main --> Harness["Game Harness"]
Main --> Preview["Loopback preview"]
Main --> Playtest["Experience evaluator"]
Main --> Environment["Godot environment manager"]
Main --> Broker["Media tool broker"]
Main --> Gate["Asset store + host attestation"]
end
Harness <-->|"JSONL · stdio"| Codex["Codex App Server"]
Codex --> Agents["Planner · Implementer · Reviewer"]
Agents --> Workspace["Web or Godot 4 workspace"]
Broker --> Providers["Image · audio · 3D providers"]
Providers --> Gate
Gate --> Workspace
Preview --> Workspace
Playtest --> Preview
Environment --> Godot["Godot 4<br/>headless import · validate · export"]
Godot --> Workspace
npm run typecheck # renderer + main process
npm test # unit and integration tests
npm run build # production renderer and main bundles
npm run verify # typecheck + tests + production build
npm run smoke:ui # isolated Electron UI screenshot
The following smoke tests use a signed-in Codex account and may consume a small amount of Codex or media-provider quota:
npm run smoke:codex
npm run smoke:harness
npm run smoke:media
npm run smoke:image
npm run smoke:model3d
npm run smoke:godot
npm run smoke:playtest
To produce an unsigned macOS DMG locally:
npm run package:mac
Public distribution still requires Developer ID signing, Apple notarization and stapling.