Spielerando

Claude Code Game Development Skills: Write Your Own

Claude Code game development skills are instruction files that Claude Code loads when your request matches them, or when you type their name. Big ready-made packs exist, but a small skill written for one chore in your own game often helps more.

What exactly is a skill?

According to the Claude Code docs on 2 November 2026, a skill is a folder holding a file named SKILL.md. Project skills live in .claude/skills/<name>/, and personal ones in ~/.claude/skills/<name>/. A description at the top tells Claude when to use it, and you can start one by hand with /name.

Unlike notes that are read every session, a skill stays out of the way until it is needed.

Should I install a game pack?

Pack What is inside Think about
Claude-Code-Game-Studios A community template with 49 agents and 72 or 74 skills (its page gives both), for Godot 4, Unity and Unreal 5, MIT licence You clone it as your project, and it needs Git and Python 3. A lot for a first game
Unity's plugin 29 engine skills at launch Only for Unity
Phaser's skills/ folder Skills for Phaser's main parts, in its repository Only for Phaser

Anthropic's help center says to install skills only from trusted sources, and its plugin docs warn that a plugin can run code on your computer with your permissions. So look before you install. A skill is mostly text, which makes that quick, but read any bundled scripts too.

Writing your own: a worked example

Marsh Warden is an invented game with a growing list of enemies. Adding one means touching four places, and every few times the AI forgets one. The new heron exists, but it never appears, because nobody put it in the spawn table.

A chore you repeat, with steps that can be forgotten, is what a skill is for:

---
name: add-enemy
description: Use when adding a new enemy type to Marsh Warden.
---

1. Draw the enemy's behaviour as a Mermaid diagram and check it for dead ends.
2. Build from that diagram: add the enemy to data/enemies.json.
3. Add it to the spawn table in data/levels.json.
4. Give it a placeholder picture and add that to art-list.md.
5. Show the closest existing enemy, with both sets of numbers side by side.

Step 1 does the most. Mermaid is a text format for diagrams, viewable at mermaid.live, and putting it in a skill turns a good intention into a habit: each new enemy gets drawn and walked through before any code exists. Skills are a strong home for that kind of rule, because they fire on the job itself.

Good candidates are adding a level, adding an item, or writing patch notes. "Make the game better" is a bad one, because it has no steps.

Skills hold your own habits. Outside knowledge fits an MCP connection better. Spielerando works that way: it gives your AI blueprints, ready-made code and a wiki, needs a free account with Google or e-mail, and does not look at your game or code (setup).

Paste this to your AI

Look at how enemies are added in this project. Find the last two enemies that were added and list every file each one touches.

Then write a project skill in .claude/skills/add-enemy/SKILL.md, using only steps you found in those files, and name the file where you saw each step. Keep it under 20 lines. A step you think is wise but did not find stays out; mention it to me separately.

Make this the skill's first step: before adding an enemy, draw how it behaves as a Mermaid diagram, such as patrol, chase, attack and stunned, check that no behaviour is a dead end, and build the enemy from that diagram.

Tell me the description you chose, and give me two requests that should trigger the skill and one that should not.

Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP

How to check it worked

  • Ask in plain words. Say "add a mud heron" without naming the skill. Claude Code should use it.
  • Start it by hand. Type /add-enemy and see that it runs.
  • Look for the heron in all four places, then play until it appears.
  • Ask for something unrelated, such as a new menu colour. The skill should stay quiet.

Sources

Last updated: 2026-11-02