How to use the Team Polish Orchestrator agent skill

What the team-polish skill makes the agent do

When you invoke the skill with a feature or area name, the agent runs a multi-phase workflow. Specialist roles take part: a performance analyst, a technical artist, a sound designer and a QA tester, and sometimes engine and tools programmers. Each examines the area, proposes polish and optimisation work, applies it, and then reports results.

The skill asks you to approve choices at phase transitions. It also reads a configuration file that decides how many specialists are active and how much review to apply. If you invoke it without an argument, it only prints usage guidance and does not spawn any agents.

• Performance analyst: profiling, memory analysis, frame budget and the optimisation list
• Technical artist: VFX polish, shader optimisation and visual quality
• Sound designer: audio polish, mixing, ambient layers and feedback sounds
• QA tester: edge case, regression and soak testing
• Engine and tools programmers: engine-level and content-pipeline fixes, used when the analysis points to them

Install the team-polish skill

Get the skill from this page as a ZIP file. You can install it in one of two ways.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: upload the ZIP under Settings → Capabilities → Skills

Check the requirements before you run it

The skill is built for Claude Code with the Claude Code Game Studios agent definitions installed. It needs the agents performance-analyst, technical-artist, sound-designer, qa-tester, engine-programmer and tools-programmer.

The skill also calls a bash helper at .claude/hooks/yaml-helper.sh and refers to two documents, .claude/docs/automation-modes.md and .claude/docs/config-resolution.md. These are not included in the file you download, so you must have them in your project for the skill to run as documented.

• Claude Code with the Claude Code Game Studios agent definitions
• The bash helper at .claude/hooks/yaml-helper.sh
• The referenced files .claude/docs/automation-modes.md and .claude/docs/config-resolution.md

Trigger the skill with a feature or area name

Invoke the skill in Claude Code with the slash command and the feature you want to polish. The argument-hint in the skill is an optional review level, which you can set to full, lean or solo. Pass a clear area name so the analysis has a defined scope.

The skill's own usage note gives examples of valid areas: combat, main menu, inventory system and level-1. Without a feature or area name, the skill outputs usage guidance only.

• Invoke with a feature: /team-polish combat
• Set review depth: /team-polish inventory system --review lean
• Without an argument: the skill prints usage and stops

Understand team size, automation and safety

The configuration decides which agents run. At the default team size, only the performance analyst and technical artist are active; other roles are consulted through them. At the larger sizes, more specialists are spawned. Before work starts, the skill states which agents will run and which will not.

In collaborative mode you must approve each phase before the next begins. In guided mode the pipeline advances unless a phase is blocked. In autonomous mode it runs end to end and records each phase outcome. Autonomous mode can edit files and may apply changes automatically, so only use it on a project you can review and restore.

The skill runs scripts, because it calls a bash helper. It does not use the network. Its tools include reading, searching, writing and editing files, so review the changes it makes.

• Default team size runs only performance-analyst and technical-artist
• Collaborative mode requires your approval at each phase
• Autonomous mode can edit files without stopping for review
• The skill runs a bash script but does not use the network
• Commit or back up your project before running it

Get the most from a polish run

Give the skill one feature at a time. A narrow scope gives the specialists something concrete to measure and keeps the token cost down. Because the pipeline spawns several subagents, it is more expensive than a quick fix, so save it for passes where profiling and edge-case testing matter.

Read the agent's analysis before you approve a phase. The decision prompts are where you control what changes are made.

• Pick one feature, such as a single level or menu, per run
• Use it for a planned pass before release rather than a one-line bug fix
• Keep your project under version control so you can review and revert edits