How to use the Guided Onboarding agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a certain kind of task. This skill, from the Donchitos/Claude-Code-Game-Studios repository (MIT licence), turns the agent into a guide for someone new to the project.

When it runs, the agent reads the project files and asks where you are starting from: no idea, a vague idea, a clear concept, or existing work. It then states what it assumes about your project and names one next step, such as /brainstorm or /setup-engine. It does not run that step itself, so you stay in control of what happens next.

• Reads the project files before asking anything
• Asks where you are starting from
• Shows its assumptions about the project
• Names one next step and does not run it
• Can save a few project settings (stage, rigor, automation mode), but only after you explicitly agree

What you need before installing

The skill is written for Claude Code or a compatible SKILL.md agent. It needs Bash available so it can run the helper scripts in .claude/scripts/ (project-state.sh, set-stage.sh and config-set.sh). It also depends on the rest of the Claude Code Game Studios template, including its other skills, docs and scripts, so installing this file alone is not enough.

The skill assumes game development and the studio's game-design files and commands. It is not meant for projects outside game development.

• Claude Code or a compatible SKILL.md agent
• Bash available for the helper scripts in .claude/scripts/
• The full Claude Code Game Studios template

How do you install the start skill?

You can install the skill for yourself or for a single project. Download the ZIP of the skill from the directory page and unzip it. Put the resulting folder in ~/.claude/skills/ to make it available to you everywhere, or in .claude/skills/ inside a project to limit it to that project.

If you use the Claude apps instead of the file system, you can upload the skill under Settings, then Capabilities, then Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into .claude/skills/ inside your project folder
• Claude apps: Settings, then Capabilities, then Skills, then upload the skill

How do you trigger the skill with a natural request?

You do not need a special command. The skill is meant to start when you describe your situation in ordinary words. Its description says it applies when you are new, have just cloned or installed the studio, have no idea or engine yet, ask where to begin, or want to bring in an existing game.

Requests that match this situation include the examples below. Say what you have in your own words; the agent will ask the rest.

• I just cloned this studio template. Where do I start?
• I have no idea what game to make yet.
• I have a rough idea for a farming game. What should I do first?
• I already have a prototype and some code. How do I bring it into this workflow?
• I'm new to this. Can you walk me through setting things up?

What happens during a session?

The agent first runs bash .claude/scripts/project-state.sh to read the project's engine, code, concept, design documents and existing work. It keeps these facts to check your answer, and does not dump them at you.

It then asks: where are you starting from with your game? You choose A (no idea yet), B (vague idea), C (clear concept) or D (existing work). If your answer conflicts with what it found on disk, it says so once and asks again.

Based on your answer, the agent names one next step. Before saving anything, it shows a short card with the stage and rigor it plans to set, and asks whether to save them to project.yaml. You can decline, and nothing is written.

• Phase 1: read the project (no questions asked yet)
• Phase 2: ask where you are starting from
• Phase 3: name the immediate next step
• Phase 4: show assumptions and ask before saving
• Phase 5: ask how much to confirm as you work, if no automation mode is set

What does the skill change in your project?

The skill writes only three settings, and only after you say yes: project.stage (with its legacy mirror production/stage.txt), modes.rigor and modes.automation. These are written through helper scripts, not by editing files by hand.

It never writes modes.review_mode or production/review-mode.txt, and it does not write the other settings that rigor affects. Rigor sets six other settings; /settings shows them and lets you change any of them at any time.

If you are already past onboarding, the skill asks nothing. It reports your engine, concept and review mode, suggests /help for the next step, and stops.

• Writes only after an explicit yes
• Writes project.stage and production/stage.txt, modes.rigor and modes.automation
• Never writes modes.review_mode or production/review-mode.txt
• Returning users with an engine and concept are not asked anything

What about safety: scripts and network access?

The skill runs scripts. Its allowed tools include Bash limited to the three helper scripts in .claude/scripts/. It does not use the network. Read the scripts before you run the skill if you want to confirm what they do.

The skill also notes a remote check. If the clone's git origin still points at donchitos/claude-code-game-studios, a push or pull request would go to the template repository rather than your own. A fork under another owner is treated as your own.

• Runs scripts: yes, limited to the helpers in .claude/scripts/
• Uses network: no
• Warns if the git origin points at the template repository

Tips for getting the most from it

Describe your idea in a few words and let the agent suggest the rest. A short, honest answer to the first question works better than a long guess.

If the agent's view of your project looks wrong, say so. If it finds code but you chose no code, or it finds no design documents, it will ask again rather than assume.

Use /settings later to change the stage, rigor or other modes without rerunning the skill.

• Start with a simple description, even if it is vague
• Correct the agent if its reading of your files is wrong
• Use /settings to change saved modes later