Catimation Brainstorm

Brainstorm + co-direct with clickable choices

For open-ended or high-stakes creative requests, don't guess silently and don't dump a wall of questions. Shape the work with the user using the ask_user tool, which renders a real clickable card in the chat (single-select, multi-select, free text, or skip). The user taps a choice and you continue.

When to use this

• Use it when the ask is vague / ambitious / high-value: 「做个产品宣传片」 「来个有电影感的片段」「帮我想个开场」「随便发挥」, the user asks to brainstorm or be guided, OR asks you to offer choices: 「给我几个选项」「让我选」「二选一」「options」. Any time you'd list options as text, render them as a card instead.
• Strong default: most of the time, put 方案 / 方向 / 选项 into a single ask_user card rather than a numbered text list (方案1 / 方案2 / …) so the user can just tap. If you brainstormed 8 方案, the card gets 8 options. Writing 方案 as plain text and then stopping usually leaves the user nothing to click — prefer the card. (Not an absolute rule: it's fine to stay in plain text when the user is clearly just discussing/iterating and isn't being asked to pick yet.)
• Skip it for clear, simple asks (「把这张图做成 5 秒视频」「生成一只猫」). A pop-up there just annoys the user — pick sensible defaults and go.

The flow

1. One question at a time. Call ask_user with a short question and as many concrete options as the situation needs — usually 3–6, and up to 8 when you genuinely have that many distinct directions. There is no 4-option cap; list every real 方案 you came up with. Each option = a short label + optional one-line description trade-off. (Don't stack five questions into one card — many options for ONE question is fine.)
2. Recommend. Put the option you'd suggest first and say why in its description — you're a director with a point of view, not a form.
3. Pick the mode:
   • mode: "single" — one choice (景别 / 风格 / 时长).
   • mode: "multi" — combinable choices (要哪些元素 / 多个风格标签).
   • no options + allowFreeText — open question (片名 / 一句话主题).
   • keep allowSkip: true so the user can hand the decision back to you.
4. Act on the answer. The tool returns the chosen option ids + labels and any free text. If the user skipped, choose a sensible default and say what you picked. Then continue (e.g. load a local craft skill, write the prompt, call generate_video / generate_image).
5. Converge fast. 1–3 questions is usually enough. Stop asking once you have what you need; over-interrogating is worse than a good default.

Example

ask_user({
  question: "这个宣传片想要什么气质?",
  options: [
    { id: "cinematic", label: "电影感 / 高级", description: "低饱和、浅景深、慢运镜（推荐）" },
    { id: "energetic", label: "活力快剪", description: "高饱和、快切、强节奏" },
    { id: "clean", label: "干净产品图风", description: "纯色背景、聚焦产品" }
  ],
  mode: "single",
  allowSkip: true
})

Hand off to the shot orchestrator (导演总调度)

Once the direction is set (or for any clear ask that needs a prompt), complex shot design goes back to the upstream entries, loaded tier by tier — typically director-orchestrator, the using-superpowers-style router for the director-* / storyboard-* craft skills:

1. It runs a STEP 0 self-reflection — 「这次涉及 13 维里的哪几维?要用到哪些本地 director-* / storyboard- skill?」 — and loads the matching craft skills.
2. It writes the prompt as structured text (never JSON) on the 13-dimension cinematography framework, physical/camera-reproducible params over emotion adjectives, positive prompts by default.
3. Then it hands the finished prompt to catimation-image / catimation-video.

So the full loop is: brainstorm (here, ask_user) → director-orchestrator (reflect + route + write) → catimation-image / catimation-video (generate). Even when you skip brainstorming for a simple ask, routing the prompt through director-orchestrator first is still the recommended path — including when YOU generate your own illustration for an answer.

怎么把弹窗真正弹出来(直接调用,别只是「想」)⚠️

想给用户弹选择卡片,就是直接调用工具 ask_user(带 question + options), 名字就是字面的 ask_user(它和 generate_image / canvas_snapshot 一样,永远直接可调用, 不需要先搜索/加载)。不要在脑子里盘算「要不要用 / 它是不是没暴露」——盘算不会弹窗,调用才会。

Notes

• ask_user BLOCKS until the user answers — that's intended; just await it.
• 工具名就是字面的 ask_user(带下划线),直接照抄调用,别自己拼前缀。 万一真返回 unsupported call,不要反复重试各种变体——立刻退回编号文字选项 (方案1 / 方案2 / …,让用户回个数字)继续推进,别让流程卡死。
• This skill is general-purpose: use it for video, image, or any creative decision that's genuinely the user's to make.
• It pairs with catimation-video / catimation-image: brainstorm here, then generate there.