skill

Tiny-Studio

An indie game studio LLM framework.

Unveil-gg/Tiny-Studiocommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/Unveil-gg/Tiny-Studio itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 26, 2026

An indie game studio in your terminal or editor.

A Claude Code-, Cursor-, and CLI-friendly template for solo and tiny teams who want AI help that feels like three talented friends in a small studio — not a AAA corporation. Workflows live in .claude/skills/; Cursor loads them through .cursor/skills, a symlink to that folder. If the link is wrong after clone, or skills never appear in Cursor, see TROUBLESHOOTING.md (common causes: Windows Git checkout without symlinks, OneDrive, or missing Developer Mode / elevated shell for mklink).

Highlights

Three agents, many skills — developer, designer, and artist. Each writes a spec file and stops. Workflows run from /start through /vertical-slice.
Client-flexible, engine-agnostic — use Claude Code, Cursor, or Codex CLI; no required src/ layout or engine stack in the template.
Optional vision MCP — screen snapshots (tiny-vision) for playtest evidence. Audio, image, and mesh generation stay in an external tool.
Single source for skills — edit .claude/skills/ only; Cursor stays in sync via .cursor/skills when the symlink is intact (TROUBLESHOOTING.md if not).

What this is

CLAUDE.md — master operating guide for the repo.
AGENTS.md — thin bootstrap for Codex CLI (and other tools that read AGENTS.md); full detail stays in CLAUDE.md.
.claude/agents/ — exactly three agents: game-developer, game-designer, game-artist.
.claude/skills/ — slash workflows (/start, /brainstorm, …).
.cursor/skills — symlink to .claude/skills/ so Cursor stays in sync (Cursor only — see TROUBLESHOOTING.md).
.claude/docs/ — philosophy, collaboration, QA-evidence, and optional MCP setup:
- vision-setup.md — tiny-vision
- external-assets.md — slot metadata only
.claude/settings.json — starter permission hints you can extend.

There is no mandatory src/, engine, or engine-specific stack — add your game where you like; the template stays lightweight.

Quick start

Copy this folder or use it as a GitHub template (when published).
Open the project in Claude Code, Cursor, or Codex CLI.
- Cursor: project skills come from .cursor/skills → .claude/skills (verify the symlink in TROUBLESHOOTING.md, especially on Windows).
- Codex CLI: reads AGENTS.md → CLAUDE.md → skill files. Agent Skills live under .agents/skills (docs); copy or symlink .claude/skills/<name> there to reuse.
Run /start to write the design/gdd.md index and design/specs/mechanics.md.
When building gameplay: /brainstorm, /proof-of-fun, /design-feature, /implement-feature in that order.
Check the spec contract with npm run validate.
Before you show a build: /qa and /ship-check.

Slash commands (skills)

Command	Purpose
`/start`	Index + mechanics; set studio mode
`/brainstorm`	Shape concepts, verbs, emotional goals
`/proof-of-fun`	30-second slice; human go / no-go
`/design-feature`	Lean feature spec (one role per turn)
`/implement-feature`	Build in slices (developer-led)
`/art-direction`	Write `design/specs/art_style.md`
`/playtest-review`	One role appends a review note, then stops
`/qa`	Evidence-first quality pass
`/ship-check`	Share/release readiness for the stated scope
`/gen-gdd`	Deepen one spec file, then stop
`/vertical-slice`	Specs → implement → `/qa`

Names may appear with or without the slash depending on your client. Skill files: .claude/skills/<name>/SKILL.md.

Optional MCP

Vision is optional. Skills work without it.

Vision — `tiny-vision`

Screen snapshots for playtest evidence. See .claude/docs/vision-setup.md.

pip install -r requirements-vision.txt
# Cursor MCP: python -m core.vision.mcp_server  (cwd = repo root)

Assets

This repo does not generate audio, images, or meshes. Slot metadata lives in design/specs/asset_hooks.json. Attach an external tool such as Mason in your client when you want binaries. See .claude/docs/external-assets.md.

Spec check

npm install
npm run validate

studio.config.json holds studio mode, turn budgets, and artifact paths. No API keys.

Invoking agents

Use your client’s subagent or Task flow with the markdown files in .claude/agents/. Each file describes voice, ownership, and collaboration.

Who it is for

Solo developers using AI for design + code + art direction conversations.
Pairs or trios who want shared vocabulary without enterprise workflow.
Anyone who liked the idea of a “game studio in a repo” but not the scale of the big templates.

Hooks

No hooks are required. Add minimal scripts under .claude/hooks/ only if they solve a real problem (e.g. your team wants a pre-commit lint). The default template stays quiet.

Customization

Keep AGENTS.md short; put durable studio rules in CLAUDE.md.
Edit CLAUDE.md with your non-negotiables (platform, engine, tone).
Tighten or loosen agent prompts in .claude/agents/.
Add one skill at a time under .claude/skills/ — keep the set small.

License

MIT — see LICENSE.