| name | game-design |
| version | 2.1.0 |
| description | Repository-aware game design diagnosis, decisions, and validation. |
| agent | ui-design-engineer |
| user-invocable | true |
| allowed-tools | ["Read","Write","Edit","Bash","Grep","Glob"] |
| routing | {"triggers":["game design","improve this game","game improvement","game design audit","game design report","core loop","game feel","player motivation","game balance","game economy","game onboarding","first-time experience","game pitch","game design document","game prototype","game scope","game progression","game fairness","game diagnostic","retention","churn","engagement"],"not_for":"Implementing a game in Phaser or Three.js, generating game art, or QA automation without a design question.","pairs_with":["game-pipeline","phaser-gamedev","threejs-builder","decision-helper"],"complexity":"Medium","category":"game-design"} |
Game design
Convert a concept, game repository, playable build, player finding, or design document into a professional, evidence-led design decision. This skill carries a complete original reconstruction of the assessed game-design capability set; use its references as operational expertise, not as a menu of shallow lenses.
Discovery and help mode
When the request is bare game design, asks what game-design help is available, or asks which review to run, read references/capability-catalog.md. Present the complete domain-organized catalog, offer the packet(s) that match the stated player moment, and state that full audit, health check, or design report runs the all-packet repository diagnostic. Do not return a partial topical menu: the catalog is the user-facing inventory of all 61 runnable capabilities.
Autonomous improvement mode
When asked to improve a game, its retention, churn, engagement, or player experience, read references/autonomous-improvement.md and follow it as the default operating mode. This is a greedy, repo-first improvement cycle: inspect the real game, run every relevant capability (all 61 for systemic retention or whole-game requests), make the smallest safe and reversible improvement that evidence supports, verify it, and leave a measurement plan for the next cycle. Never wait for a feature request when the evidence itself identifies a material player harm or opportunity.
1. Intake and deterministic evidence inventory
Start from repository evidence. Read the target repository's governing instruction files first. Search its installed skills and agents for game, product, UI, implementation, analytics, and research guidance; load every applicable local instruction and record its authority before drawing conclusions. Then use file search and code inspection to find design documents, player-facing copy, rules and state, UI, configuration and tables, tests, analytics schemas, issues, ownership, and recent changes. Separate facts into observed, documented, measured, and inferred.
Ask only what the repository cannot answer:
- Which player context and concrete play moment matter?
- What external player evidence exists: playtest recordings, support patterns, telemetry, reviews, or community reports?
- Which constraints are binding: platform, release phase, team, accessibility, legal, trust, time, or cost?
- Does the user need a diagnosis, options, specification, priority decision, prototype plan, or full report?
2. Greedy reference routing
Load every module that could materially change the recommendation, its player-risk assessment, or validation plan. Do not stop at the smallest topical match. Add adjacent modules when the player path crosses their domain.
