Guide · Agent skill
How to use the Team Level Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Team Level is an agent skill for Claude Code-style coding agents that turns a request like "design the forest temple" into a structured level plan. It acts as the lead of a small game-design team, reads your existing design documents, and hands parts of the job to specialist sub-agents. This guide explains what the skill does, how to install it, how to trigger it, and what it needs before you run it.
Team Level OrchestratorDonchitos/Claude-Code-Game-Studios · MIT Download the skill Quick start
- 1Download the ZIP. Download the team-level skill ZIP from this page. Keep the folder name team-level.
- 2Place the skill folder. Unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. You can also upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Prepare your design documents. Make sure your game concept, pillars, and any level or narrative docs exist in the design/ folders the skill reads.
- 4Check the wider repo files. Confirm the hooks/yaml-helper.sh script and the .claude/docs files are present, since the skill does not include them.
- 5Run the skill with a level name. Invoke /team-level with a level or area name, such as forest temple, and an optional --review value.
- 6Answer the decision points. Pick from the sub-agents' proposals at each prompt, or let the pipeline advance if you chose guided or autonomous mode.
What the team-level skill makes the agent do
When you ask the agent to design a level or area, the skill has the agent act as the orchestrator for a level team. It reads your game concept, your game pillars, and any existing level or narrative documents, then spawns specialist sub-agents for the parts of the work they cover. The named roles are level-designer, narrative-director, world-builder, art-director, systems-designer, qa-tester, and accessibility-specialist.
At each step transition the skill uses AskUserQuestion to show you the sub-agent's proposals as selectable options, so you choose the direction rather than receiving one finished output.
- Reads design/gdd/game-concept.md (or design/game-brief.md at minimal rigor), design/gdd/game-pillars.md, design/levels/, and design/narrative/ where they exist.
- Spawns sub-agents for layout, narrative, lore, visual direction, systems, accessibility, and test cases, depending on the team size setting.
- Writes working files under production/levels/[level-name]/ and test cases under production/qa/test-cases/.
Install the team-level skill from the ZIP
Download the team-level ZIP from this page. Then add the skill folder to one of two locations: your user-level skills folder at ~/.claude/skills/, or a single project's .claude/skills/ folder if you want it only for that project.
If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills. The skill folder you install contains SKILL.md only.
- User-wide: ~/.claude/skills/team-level
- Project-only: .claude/skills/team-level inside your project
- Claude apps: Settings → Capabilities → Skills
Trigger the skill with a request
Invoke the skill with a level or area name. The argument is the target you want designed. The skill also accepts an optional --review flag with the values full, lean, or solo. The flag overrides the review_mode setting in your config.
If you run the skill with no argument, it prints usage guidance and stops without spawning agents or reading any design files. Examples from the skill's own guidance include forest temple, tutorial village, and final boss arena.
- Argument: the level or area to design
- Optional: --review full, --review lean, or --review solo
- Examples: forest temple, tutorial village, final boss arena
/team-level [level name or area to design] [--review full|lean|solo]Choose a team size and an automation mode
The team.size setting decides which agents are active. The default, individual, runs only level-designer, and other agents are consulted through it. The small setting adds systems-designer, art-director, and qa-tester. The studio setting adds narrative-director, world-builder, and accessibility-specialist to the full pipeline.
The automation setting controls how decision points are handled. In collaborative mode you must approve each step before the pipeline moves on. In guided mode the pipeline advances automatically unless a step is BLOCKED. In autonomous mode it runs end to end and records each step outcome with log_decision. Some decision categories always prompt you, whatever the mode.
- individual (default): level-designer only
- small: adds systems-designer, art-director, qa-tester
- studio: adds narrative-director, world-builder, accessibility-specialist
- collaborative: you approve every step
- guided: advances automatically unless BLOCKED
- autonomous: runs end to end, logs each outcome
Understand the setup requirements and safety notes
The skill's config step runs yaml-helper.sh from the wider repository through a bash shell. Its only permitted shell call is that resolve_config command. The skill makes no network use, according to the review.
The skill depends on files outside its own folder. The hooks/yaml-helper.sh script, the .claude/docs files, and the other agent definitions are referenced but are not included in the skill folder. Without them, the skill has little to run on, so treat it as part of a wider project setup rather than a standalone tool.
- Runs a script: yes (yaml-helper.sh via bash, resolve_config only)
- Uses network: no
- Needs a bash shell and Claude Code or a compatible SKILL.md runtime
- Needs hooks/yaml-helper.sh, .claude/docs files, and agent definitions from the wider repo
Tips for better level plans
Keep your design documents current before you run the skill. The skill works from the documents it finds, so a thin design/ folder produces thin output. Start with the individual setting to check the flow, then move to small or studio when you need more specialist input.
Use collaborative mode when you want to review each proposal. Use autonomous mode only when you trust the documents and the settings, because it records decisions rather than stopping for your approval.
- Write the game concept and pillars first, since they are read at the start.
- Name the level clearly, because the name sets the output folder under production/levels/.
- Compare the output of individual and studio runs on the same level to see what the extra agents add.
What people use it for
- Plan a tutorial village with layout, pacing, and encounter notes drawn from your existing design documents.
- Draft a final boss arena with narrative purpose and visual direction, using the studio team size.
- Produce structured level docs in production/levels/ for a team that keeps its design notes in design/.
- Generate test cases for a finished level through the qa-tester role at the small or studio team size.
Frequently asked questions
Is the team-level skill free to use?
The skill is released under the MIT licence, as listed on this page. Check the licence file in the source repository for the full terms before you reuse or redistribute it.
What do I need to run the team-level skill?
You need Claude Code or a compatible SKILL.md agent runtime, and a bash shell because the config step runs yaml-helper.sh. You also need the hooks/yaml-helper.sh script, .claude/docs files, and other agent definitions from the wider repo, which are not in the skill folder.
Where do I install the team-level skill?
Unzip the ZIP into ~/.claude/skills/ for all projects or into a project's .claude/skills/ folder for one project. In the Claude apps, upload it under Settings → Capabilities → Skills.
Why does the skill say it needs design files?
The skill reads your game concept, pillars, and level or narrative docs before it spawns any agents. Without a design/ folder it has little to work from, so the output will be limited.
How does the team-level skill differ from asking an agent for a level idea directly?
The skill runs a multi-step pipeline with sub-agents and approval points, which takes more time and uses more resources than a one-off idea. The source describes it as suited to structured level docs rather than quick brainstorming.
What should I do if the skill does nothing when I run it?
Check that you passed a level or area name, since running it with no argument only prints usage guidance. Also confirm that the hooks/yaml-helper.sh script and a bash shell are available, because the config step depends on them.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use team-level if you are building a game level pipeline with an AI agent, already keep design documents, and want a step-by-step approval flow with structured level docs. Pick something else if you work outside games, do not have a design document set, or only want a quick one-off level idea.
Download the skill