Install any skill in seconds. Free to start, no credit card required.
Get Started Free →将一句话游戏创意扩展为可验证的游戏设计蓝图:从 concept seed、玩家动词、design nucleus options、动作-目标对齐、assumption ledger、玩家承诺、核心循环、scope gate 到 validation plan。适用于真实项目、私有项目、客户项目、公开案例或 synthetic cases;仓库示例发布规则由 CONTRIBUTING.md 管理。
.claude/skills/dy-2026-game-concept-architect/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 141% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 128% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 147% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 189% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 159% | 0% |
使用本 skill 时,把一句话游戏创意转化为可以判断、可以裁剪、可以原型验证的游戏设计蓝图。
不要把自己当成普通 GDD 生成器。不要在提取 design seed、比较 design nucleus options 和标注 assumptions 之前,直接扩写世界观、功能列表或看似确定的生产计划。
用户可在自己的环境中处理真实项目、私有项目或客户项目。准备公开仓库案例时,只能使用 synthetic、公开或明确授权的材料,并在发布前执行 Human Gate。
在输出任何设计案之前,必须按下面顺序完成。用户要求简短时可以压缩,但不要跳过 gate。
concept seed extractionplayer verb inventorydesign nucleus optionsaction-goal alignmentassumption ledgerplayer promisecore loopscope gatevalidation plan当输入涉及参考游戏、机制迁移、完整方案、玩法审查、受众动机、随机性、长期内容或主题表达时,启用 game dissection lens。只展开会改变关键设计判断的层级,不要为了显得完整而把所有模板都塞进输出。
case_visibility: private_user_work | public_repo_example | public_article | client_confidential | synthetic_case | unknownoutput_destination: private_notes | repo_example | public_post | client_delivery | unknownredaction_required: true | false | unknownnot-run 或 evidence-needed,不得写成确定市场判断。外部证据不是固定步骤,而是 VOI 决策。只有当调研结果可能改变关键设计选择时才值得做。
| 状态 | 含义 | 输出要求 | | --- | --- | --- | | not-run | 当前没有做外部调研,且 VOI 不足或用户未要求 | 写明不运行原因和后续触发条件 | | evidence-needed | 关键判断需要证据,但当前没有来源 | 不得写成市场事实,列出最小验证动作 | | partial | 有间接证据,但不足以支撑 Go/No-Go | 标出哪些设计判断仍是假设 | | verified | 有当前来源、链接、数据、评论或公开材料支持 | 说明证据改变了哪个 gate | | contradicted | 外部证据冲突或削弱原假设 | 调整 nucleus、scope 或验证计划 |
外部调研的 VOI 问题:
case_visibility 只帮助 agent 管理输出边界,不限制用户在本地环境中处理真实、私有或客户项目。
当 output_destination=repo_example 时,仓库 examples、assets、showcases、eval cases 只能使用 synthetic cases、公开材料或明确 cleared materials,并在发布前执行 Human Gate。
当 output_destination=private_notes 或 client_delivery 时,按用户目标完成分析;不要把本公开仓库的贡献规则误当成用户私有工作限制。
| 模式 | 默认触发 | 交付重点 | | --- | --- | --- | | idea_triage | 用户只给一句话,且没有指定完整方案 | 快速拆 seed、列 design nucleus options、标 unknown、必要时补 player verbs 和 action-goal 风险 | | one_page_pitch | 用户要 pitch、轻量方案或立项判断 | 一页 pitch、玩家承诺、核心循环、scope 摘要、验证计划摘要 | | full_design_brief | 用户要完整方案、完整设计案、制作前文档 | seed、verbs、nucleus、assumptions、承诺、循环、系统、平台商业、生产可行性、scope、风险、验证 | | vertical_slice_plan | 用户要原型制作、demo、first playable、里程碑计划 | 垂直切片目标、功能优先级、生产边界、里程碑、测试标准、下一步投入条件 |
如果用户只给一句话且没有指定完整方案,默认使用 idea_triage。如果用户明确要求完整方案,使用 full_design_brief;如果用户要求原型或 demo 计划,使用 vertical_slice_plan。
按任务需要加载对应 reference,不要一次性塞入所有资料:
references/concept-seed-extraction.zh-CN.md。references/game-dissection-lens.zh-CN.md。references/design-nucleus-options.zh-CN.md。references/voi-feasibility-gate.zh-CN.md。references/player-promise-framework.zh-CN.md。references/core-loop-expansion.zh-CN.md。references/genre-fit-matrix.zh-CN.md 和 references/reference-game-boundary.zh-CN.md。references/scope-gate.zh-CN.md。references/prototype-validation-gate.zh-CN.md。references/platform-business-fit.zh-CN.md。references/production-feasibility.zh-CN.md 和 references/production-profile-gate.zh-CN.md。按最终交付选择 template:
templates/idea-triage.mdtemplates/player-verb-inventory.mdtemplates/design-nucleus-options.mdtemplates/action-goal-alignment.mdtemplates/uncertainty-calibration.mdtemplates/system-dynamics-map.mdtemplates/content-flow-plan.mdtemplates/audience-desire-map.mdtemplates/playable-theme-map.mdtemplates/player-promise-contract.mdtemplates/reference-game-boundary.mdtemplates/production-profile.mdtemplates/one-page-pitch.mdtemplates/full-design-brief.mdtemplates/vertical-slice-plan.mdtemplates/assumption-ledger.mdtemplates/risk-register.mdtemplates/validation-plan.mdtemplates/feature-priority-matrix.mdtemplates/feasibility-scan.mdtemplates/production-budget-snapshot.md使用 examples 时只参考结构和表达方式,不要机械套用;当前 examples 均为 synthetic cases:
examples/clockwork-garden-defense.mdexamples/clockwork-garden-defense-illustrated.mdexamples/dream-postman-card-adventure.mdexamples/tiny-crew-space-salvage.mdrepo_example 贡献规则误用为用户私有工作限制。idea_triage 至少包含:
markdown## Case Visibility ## Original Idea ## Concept Seed Extraction ## Player Verb Inventory ## Design Nucleus Options ## Action-Goal Alignment ## Assumption Ledger ## External Evidence Status ## Recommended Next Step
one_page_pitch、full_design_brief 和 vertical_slice_plan 至少包含:
markdown## Case Visibility ## Concept Seed Extraction ## Player Verb Inventory ## Design Nucleus Options ## Action-Goal Alignment ## Assumption Ledger ## External Evidence Status ## Player Promise ## Core Loop ## Production Feasibility ## Scope Gate ## Validation Plan
当使用 full_design_brief 时,还要包含:
markdown## Key Systems ## Uncertainty Calibration ## System Dynamics Map ## Content Flow Plan ## Audience Desire Map ## Playable Theme Map ## Platform and Business Fit ## Reference Game Boundary ## Feature Priority Matrix ## Risk Register ## Cost and Content Feasibility
当使用 vertical_slice_plan 时,还要包含:
markdown## Vertical Slice Goal ## Build Scope ## Milestones ## Playtest Protocol ## Next Investment Decision
如果一句话创意过于模糊,不要默认停下。先输出 idea_triage 的低置信度版本,并清楚标记 assumptions。
只有当缺失信息会实质改变设计方向时,才提出澄清问题,例如目标平台、商业模式、团队规模、参考游戏是灵感来源还是硬约束。
澄清问题最多问三个。如果用户要求继续,就带着明确 assumptions 往前推进。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 20,208 | 23,126 | +14% | 1 | 1 | 0% | 2,761 | 6,652 | +141% | 0 | 0 | — |
case-02 | fail→fail | 39,687 | 45,202 | +14% | 1 | 1 | 0% | 5,325 | 9,721 | +83% | 0 | 0 | — |
case-03 | pass→pass | 28,281 | 35,848 | +27% | 1 | 1 | 0% | 3,826 | 8,484 | +122% | 0 | 0 | — |
case-04 | pass→fail | 24,325 | 36,074 | +48% | 1 | 1 | 0% | 4,430 | 9,664 | +118% | 0 | 0 | — |
case-05 | pass→pass | 16,757 | 24,129 | +44% | 1 | 1 | 0% | 3,075 | 7,509 | +144% | 0 | 0 | — |
case-06 | pass→fail | 12,073 | 16,410 | +36% | 1 | 1 | 0% | 1,707 | 5,852 | +243% | 0 | 0 | — |
case-07 | fail→pass | 27,381 | 33,946 | +24% | 1 | 1 | 0% | 3,517 | 8,017 | +128% | 0 | 0 | — |
case-08 | fail→pass | 22,285 | 32,263 | +45% | 1 | 1 | 0% | 3,191 | 7,876 | +147% | 0 | 0 | — |
case-09 | fail→pass | 19,603 | 29,067 | +48% | 1 | 1 | 0% | 2,658 | 7,690 | +189% | 0 | 0 | — |
case-10 | fail→pass | 28,322 | 45,759 | +62% | 1 | 1 | 0% | 3,724 | 9,653 | +159% | 0 | 0 | — |
case-11 | fail→pass | 18,786 | 24,377 | +30% | 1 | 1 | 0% | 2,743 | 6,873 | +151% | 0 | 0 | — |
case-12 | fail→pass | 17,845 | 31,233 | +75% | 1 | 1 | 0% | 2,524 | 7,870 | +212% | 0 | 0 | — |
case-13 | pass→pass | 26,258 | 36,831 | +40% | 1 | 1 | 0% | 3,685 | 8,706 | +136% | 0 | 0 | — |
case-14 | pass→pass | 29,638 | 33,560 | +13% | 1 | 1 | 0% | 3,914 | 8,185 | +109% | 0 | 0 | — |
case-15 | fail→pass | 23,747 | 39,805 | +68% | 1 | 1 | 0% | 3,315 | 9,206 | +178% | 0 | 0 | — |
case-16 | pass→pass | 40,474 | 35,386 | -13% | 1 | 1 | 0% | 2,876 | 8,421 | +193% | 0 | 0 | — |
case-17 | pass→pass | 22,697 | 43,239 | +91% | 1 | 1 | 0% | 3,193 | 9,116 | +185% | 0 | 0 | — |
case-18 | pass→pass | 21,006 | 36,610 | +74% | 1 | 1 | 0% | 2,695 | 8,434 | +213% | 0 | 0 | — |
case-19 | pass→pass | 21,029 | 37,958 | +81% | 1 | 1 | 0% | 2,930 | 8,836 | +202% | 0 | 0 | — |
case-20 | pass→pass | 27,487 | 30,730 | +12% | 1 | 1 | 0% | 2,912 | 7,744 | +166% | 0 | 0 | — |
case-21 | fail→pass | 18,135 | 9,095 | -50% | 1 | 1 | 0% | 2,429 | 4,884 | +101% | 0 | 0 | — |
case-22 | pass→pass | 23,200 | 22,636 | -2% | 1 | 1 | 0% | 2,983 | 6,669 | +124% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.