Game Design

Game design skills for AI agents working on playable games.

Game Design is a skills plugin for Codex and Claude Code. It helps you design or revise game actions, puzzles, worlds, economies and content in a new or existing game. Work on one concrete rule or combine methods for a change across systems. A large game design document, score, victory condition or simulation is not required.

What you can do

Design actions, combat timing, actor behavior and encounters with gameplay design.
Build puzzles, clues, learning sequences and player assistance with information design.
Connect places, inhabitants, quests and conditional scenes with world and narrative design.
Work out economies, progression, loss, recovery and persistent state together with systems design.
Design shared play, access, leaving and returning with participation design.
Create and select encounters, items and other related game content with content design.
Reconcile cross-system decisions or adapt an existing game with game design.

The supported package is the complete collection listed in catalog.json. Choose a specialist directly for a focused task; the general game-design method is not a mandatory router. Skills link to related references when a decision crosses their boundary.

Demonstration

East Gate is a small playable puzzle about adapting a clue to a different language and display. Its Russian version presents one line at a time, lets the player pin lines, and accepts the direction encoded by their first letters. A correct choice opens a gate and supplies the next scene; a wrong direction remains a valid retry. The example consumer checks the content and action relationships, not language quality or human experience.

For tasks that need executable artifact evidence, use the optional strategy matrix, Ink episode, or glTF/GLB inspection. Their dependencies are prepared in task-owned directories; paper and prose work installs none.

Install

Use a plugin-capable Codex or Claude Code CLI. The commands below fetch the public GitHub repository through each client's native plugin manager; they can be entered in Bash or Windows PowerShell. Check the source and scope of any existing game-design-source registration before adding another one.

Codex

codex plugin marketplace add Muratovnik/gamedesign-skills
codex plugin add game-design@game-design-source
codex plugin list --marketplace game-design-source --json

Claude Code — run from your game repository so --scope local applies to that project.

claude plugin marketplace add Muratovnik/gamedesign-skills --scope local
claude plugin install game-design@game-design-source --scope local
claude plugin details game-design

Verify that the installed component inventory shows the Game Design skills, then start a new session. The GitHub-source syntax is documented by both clients, but this project's recorded native installation checks use a local source directory, not the GitHub route. See installation and scope and compatibility evidence.

For use without plugin registration, download the v0.2.0 source release or clone the repository, then give your agent the authorized source directory, game root and the applicable skills/<name>/SKILL.md. Reading methods this way needs neither Python nor a plugin manager. General research and engineering methods are maintained separately in Assay; Game Design does not install Assay. Its exact pinned revision and conditional requirements are in the consumer contract.

Quick start

After installing the plugin, start a fresh session. To try one method without preparing a game repository, send the following prompt:

Use Game Design's gameplay-design skill. In a top-down game an enemy charges for 0.6 seconds and strikes one tile. The warning appears only 0.1 seconds before impact. Revise the warning and player-response rules. Give exact timings, the available player responses, and a failure case to check. Do not change code.

The result should describe a concrete action sequence with a plausible failure case, rather than just naming design principles. This is an example request, not a tested model run or player playtest.

For an existing game, supply its current rules or files, the desired change and the actions the agent may take. Name the skill explicitly; installation does not guarantee automatic selection. See the Russian quick start.

Documentation

Installation covers GitHub installation, direct source use, scopes, updates and removal.
Compatibility records the current dependency and client-qualification boundaries.
Example guide maps methods to playable studies and tools.
Russian quick start provides the first-use route in Russian.
Design decisions and research limits explain the package's design basis.

Limits

The examples are synthetic. Deterministic and headless checks establish only their inspected properties, not player perception, arbitrary-engine compatibility or general model quality. Installed-resource loading and automatic model selection are separate claims; their recorded evidence and limits are in compatibility.

Contributing

See CONTRIBUTING.md for source ownership and maintainer checks.

License

The package is available under the MIT License. Technical plugin-schema material has the additional terms described in NOTICE.md.