gat-brainstorm
Brainstorm a game idea through one-question-at-a-time designer interviews. Produces game.md, systems-index.md, and art-direction.md, or runs as discussion-only.
Yuki001 · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add Yuki001/game-dev-skills --skill gat-brainstorm技能说明
name: gat-brainstorm description: "Brainstorm a game idea through one-question-at-a-time designer interviews. Produces game.md, systems-index.md, and art-direction.md, or runs as discussion-only." argument-hint: "[ | discuss]" user-invocable: true allowed-tools: Read, Glob, Grep, Write, Edit, Agent, AskUserQuestion
Brainstorm
This skill explores a game concept through open-ended designer interview.
Spawn gat-designer for design reasoning and gat-artist for visual direction.
Pick questions from the angle table below — follow the conversation, not a script.
Phase 1: Resolve Mode
- If argument is
discuss→ Mode:discuss(no files written, exploration only) - If argument is a hint or empty → Mode:
design(produces foundational design docs and global art direction)
Check whether gat/overview/game.md, gat/overview/systems-index.md, or
gat/overview/art-direction.md already exist. If so, note them — the interview may
refine or replace existing decisions instead of starting from scratch.
Phase 2: The Interview
Core Rules
- One question at a time. Never batch. Wait for the answer before the next question.
- Provide a recommended answer with each question. Explain the design reasoning.
- Prefer open-ended questions. Let the user type free-form responses. Reserve
AskUserQuestion(multiple-choice) for when you are presenting concrete solution options and need the user to pick one. Most of the interview should be dialogue, not forms. - Pick angles, don't follow steps. Use the angle table as a menu. Jump to whatever dimension is most useful next — chase what's interesting or ambiguous.
- If a question can be answered by reading existing design files, read them instead of asking.
- Spawn
gat-designerwhen you need a design reasoning pass: drafting a core loop, proposing systems, evaluating a trade-off, or checking consistency. - Spawn
gat-artistwhen visual identity needs synthesis: art references, palette, readability, asset groups, production standards, or conflicts between style and gameplay clarity.
Drilling into Vague Ideas
When the user has a fuzzy idea — about the whole game or a single system — your job is to make it concrete through relentless, curious questioning. This is the core of the interview.
How to drill:
- When the user says something vague ("combat should feel impactful"), ask what specifically makes it impactful — is it animation, sound, damage numbers, controller rumble, enemy reaction, time-to-kill? Keep asking until the abstraction bottoms out in concrete mechanics.
- When the user proposes a system, ask about its boundaries. What does it NOT do? What system owns the adjacent responsibility? A system without edges is still fuzzy.
- When the user describes a player experience ("I want the player to feel lost"), ask what the game does to create that feeling. What does the player see, hear, and do? What information is withheld? What mechanics produce the emotion?
- When the user references another game ("like Dark Souls but..."), isolate what exactly they want to keep and what they want to change. The reference is a shortcut — unpack it.
- Ask about edge cases and failure states. What happens when the player ignores the system? What happens when they optimize it to the extreme? The answers reveal whether the system is understood or still hazy.
- Ask about the player's moment-to-moment decisions. If the user can't describe what choices the player makes inside the system, the system isn't clear yet.
- If an answer opens three new questions, pick the most foundational one first. Resolve dependencies before details.
Signals that something is still vague and needs more drilling:
- The user uses abstract adjectives without mechanics behind them ("fun", "smooth", "deep", "cool")
- A system is named but its inputs, outputs, and rules are undefined
- Two systems have overlapping or unclear boundaries
- The user can describe what the system IS but not what the player DOES in it
- Numbers are absent where they matter (how many? how long? how much?)
Seed Extraction
If a concept hint was provided, first spawn gat-designer to extract what the
hint already answers. Briefly summarize what's established so the user can
confirm or correct before diving in. Skip if no hint.
Interview Angles
Pick questions from any angle below. There is no fixed order — follow the thread that matters most at each moment. The table is a palette, not a checklist.
| # | Angle | Purpose | Example prompts |
|---|---|---|---|
| 1 | Genre & Style | Establish the game's design identity and reference points | What genre(s) does this live in? What games should it feel like mechanically? Real-time or turn-based? 2D or 3D? Single-player, co-op, or competitive? |
| 2 | Visual Direction | Establish the global art identity that will become art-direction.md | What should the game look like at a glance? Which art references fit or should be avoided? What palette, shape language, camera, readability, and production constraints matter? |
| 3 | Core Player Verb | Pin down the primary action the player repeats | What does the player actually DO moment-to-moment — shoot, build, explore, talk, craft, steer, command? What makes that action satisfying? |
| 4 | Target Feeling | Define the emotional experience | What should the player feel during play — tension, mastery, wonder, power, relaxation, social connection, fear, curiosity? When do they feel it most? |
| 5 | Fantasy & Role | Clarify who the player is in the world | What fantasy does the game fulfill? Who is the player — hero, commander, survivor, creator, investigator, merchant? |
| 6 | Scope & Constraints | Set boundaries early | Rough scope (jam, indie, commercial)? Platform? Timeline? Team size? Content rating? Hard constraints? Any visual production constraints like pixel ar |