---
name: design-game
description: Turns a browser-game idea into a buildable design with a player decision, complete round, readable feedback and observable play checks. Use before building a small game or when a prototype has actions but no interesting choice.
license: CC0-1.0
compatibility: Designed for Claude Code project skills. Reads project files and writes design.md; building or browser testing is a separate task.
metadata:
  author: Claude Game Studio
  version: "1.0.0"
---

<objective>
Turn the user's idea into a small game someone can understand, finish and replay. Make design judgments yourself from the brief and existing project. Produce design.md before implementing code.
</objective>

<quick_start>
Read the supplied idea and the existing game, if any. Identify the player, device and desired feeling. If unspecified, use a short single-player browser round with keyboard and touch controls. Preserve explicit requirements. Ask only when a missing answer would change the core game.
</quick_start>

<process>
1. Find the player's decision. State the main verb and one choice with a cost the player can see. Explain why both choices can be reasonable. If a choice is always better, change its cost or reward. Theme alone is not a mechanic.
2. Define the complete round: initial view, controls, objective, repeated action, change in difficulty, win, loss and restart. Name what persists after restart. Make every promised input possible on the target device.
3. Connect requirements to rules. For each important phrase in the brief, write the mechanic it implies and the feedback that teaches it. Choose initial numbers with units; label them as design settings until measured. Work through one normal route and one edge case to check feasibility.
4. Cut scope to the smallest complete expression of that decision. Name the room, actors, assets and sounds actually needed. Read existing source and licensing before requesting replacements. Keep the first screen readable before adding detail.
5. Save design.md with premise, first second, verb/decision, loop, scope, look/sound and done-means checks. Add a short change table when revising an existing game. End with three observable checks covering a full round, an input/feedback edge case and restart. Run the game only when the user also asked to build or test; record what actually happened.
</process>

<pitfalls>
- A delivery game with constant speed and unlimited carrying makes returning early pointless. Consider load, expiry, hazards or limited stock; choose the cost that fits the brief.
- A sword game needs swing, hit and damage checks. Do not inherit magazine/reload checks from a shooter starter.
- Difficulty needs a readable warning and a possible response. An invisible timer or unavoidable hit does not teach a choice.
- Verify whether the round can actually be won before tuning its presentation. Preserve terminal states until restart.
</pitfalls>

<success_criteria>
The design preserves the user's idea, specifies an observable player decision and contains a complete playable round. Every number is a proposed setting or has an identified measurement. Existing work and licence constraints are respected. No fabricated playtest, user response, model cost or promised result. Report unrun checks plainly. Do not install packages, spend, publish or change permissions for a design-only request.
</success_criteria>
