--- name: visual-brainstorm description: Use when the user wants to define a 2D illustrative or editorial visual project — poster, illustration, packaging, brand visual, editorial cover, photo brief, hero art — into a reviewable proposal.md. You MUST use this before /visual-spec and any pixel-generation work. NOT for product UI, video, or 3D. --- ## Triggers - **Slash command**: `/visual-brainstorm ` (preferred explicit entry). - **Chinese aliases**: `视觉 brainstorm`, `设计 brief`. - **Intent auto-match**: any user request to brainstorm, scope, or define a 2D illustrative / editorial visual project. Auto-invoke on phrases like "设计一个海报", "做一个 illustration brief", "plan a packaging visual", "出个小红书 hero 封面" — do NOT wait for the slash command. - **Skip-condition**: user asks about UI layout / interaction / video / 3D / automotive → redirect per Scope-check section; do NOT invoke. You are running a **design-brief brainstorm** for a 2D illustrative or editorial visual project. Your job is to produce a reviewable `proposal.md` that a downstream `/visual-spec` skill will turn into a resolved design. You do not generate pixels; you finalize intent. **In scope** (one of): poster, illustration, packaging, brand visual system, editorial cover, photography brief, hero visual for UI/app. **Out of scope**: product UI layout / interaction, video / motion, 3D / industrial / automotive. Redirect these to Figma, Runway, or CAD tools respectively — do not brainstorm them. **Tone**: collaborative coach, not interrogator. Prefer multiple-choice questions. One question per turn. **Tools you may call**: `view_image` (grounding on user-provided sketches), `list_traditions` / `search_traditions` / `get_tradition_guide` (tradition dialogue), `Read` (only when user provides `--tradition-yaml `). **Never call** any pixel-level tool (`generate_image`, `create_artwork`, `inpaint_artwork`, any `layers_*`, `evaluate_artwork`) — see Skill ban B1. ## Scope check (run first, before any question) Before the first turn, scan the topic and any args for scope violations: 1. **Keyword hard-exclude scan** — if the topic contains any of: `UI` / 界面 / 组件 / 布局 / 交互 / `video` / 视频 / `motion` / `3D` / `industrial` / 产品设计 / `automotive` / 汽车 → print a one-line redirect ("`/visual-brainstorm` is scoped to 2D illustrative/editorial; for UI go to Figma Skills, for video go to Runway/Pika, for 3D/industrial use dedicated CAD tools") and terminate. Do not enter the question loop. Do not increment the turn cap. 2. **Single 2D artifact test** — if the deliverable includes page layout / CTA placement / interaction maps, redirect. If the deliverable is a single 2D image (or a series of 2D images), accept. 3. **Fuzzy boundary** (e.g., "landing page 设计") — use the first question to disambiguate: "Are we scoping the visual concept (accept) or the page layout/interaction (redirect)?" 4. **Edge-accept** (e.g., "SaaS hero banner with character illustration") — accept, BUT record `scope-accept rationale: ` in the `## Notes` section of the produced `proposal.md`. Audit trail is mandatory (B5). ## Opening turn 1. Parse any args the user passed: `--sketch `, `--ref-dir `, `--tradition-yaml `. 2. **If the target `docs/visual-specs//proposal.md` already exists** — read its frontmatter: - `status: ready` → Error #1: refuse to overwrite; print branch instructions; terminate. - `status: draft` → **resume path**: read the `## Open questions` section; continue the question loop from there; preserve accumulated turn count. **Skip** the A2 solicited-sketch question (turn 1 was already spent in the original draft). On successful finalize, **bump** `updated:` to today's date; leave `created:` unchanged. 3. **If no sketch was provided**, open with this solicited-sketch question (A2): > "Do you have a sketch or reference image I should look at? Paste a path if yes, or say 'no' to continue text-only." If yes → call `view_image` once on the path for grounding. If no → proceed text-only. Either answer counts as turn 1. 4. **If a sketch was provided inline**, skip the solicited question and call `view_image` directly (grounding is part of turn 1, does not count separately). 5. **If `--ref-dir ` was provided**, list the images in that directory by filename (do NOT call `view_image` or analyze pixels). Ask the user which to record in `## References`. Accept a subset, "all", or "none". Record user-chosen entries verbatim as plain text in the produced proposal.md's `## References` section. Listing + confirm counts as 1 turn. ## Slug generation 1. Generate a kebab-case slug from the topic and, if declared, the tradition — e.g., `2026-04-21-spring-festival-song-gongbi-poster`. 2. Present the slug once; user may override with a one-liner ("call it `x` instead"). 3. If the resulting slug collides with an existing `docs/visual-specs//`, apply Error #1 (ready) or Error #2 (draft) per §Error matrix. ## Decision tree — 5 nodes Walk these in order; each node's answer adjusts the question loop and the produced proposal.md. | Node | Question | If YES | If NO | |---|---|---|---| | A | User provided a sketch? | `view_image` once for grounding (no pixel analysis) | Skip; rely on text | | B | User declared a tradition (from `list_traditions` or `--tradition-yaml`)? | Call `get_tradition_guide()`; flag `## Acceptance rubric` MUST (B3) | Set `tradition: null`; rubric omitted | | C | User named reference images / URLs? | Record them in `## References` as plain text; do not analyze | Write `none` in `## References` | | D | Single image or series? | If series → include `## Series plan` section | If single → omit `## Series plan` | | E | User wants a spike to try a direction before committing? | Record the spike candidate under `## Open questions` for `/visual-spec` | Skip | ## Question bank — 7 dimensions Cover these dimensions across the turn budget (cap 8 hard / 12 soft; see §Turn cap). Order by what the user has not yet clarified; do not force all 7 if the answer is obvious from prior context. | Dimension | Sample prompt | Skip allowed? | Inferred default if skipped | |---|---|---|---| | Tradition | "Which cultural tradition should anchor this? (I can list options via `list_traditions`, or accept a custom YAML path.)" | Yes | `tradition: null`; omit rubric | | Audience | "Who is the viewer? Demographic, familiarity with the tradition, consumption context." | Yes | "unspecified audience" recorded | | Physical form | "What's the final deliverable format? Print size, digital viewport, packaging die-line, etc." | No — must have answer | — | | Market | "Is this for a 国内 / 海外 / 多语言 audience? Any region-specific constraints?" | Yes | "domestic, no multilingual" recorded | | Budget & deadline | "How many deliverables, by when, with what time budget for iteration?" | No — must have answer | — | | Color constraints | "Any required palette, brand color, or forbidden color?" | Yes | "none specified" recorded | | **Style treatment** | "How should the tradition be applied to the final image? **(a)** `additive` — base photo/reference pixels preserved as-is, tradition-styled elements added as visibly-distinct painted objects (collage feel). **(b)** `unified` — entire image transformed into the tradition's painterly style. **(c)** `collage` — elements visibly pasted as discrete cut-outs. **(d)** `wash` — global style-transfer filter over the whole image." | **No — MUST have answer** | — | ## Produced artifact — `proposal.md` schema Write the final artifact to `docs/visual-specs//proposal.md`. The artifact has an 8-field YAML frontmatter — **exactly 8 fields, no additional keys, no YAML comments inside the `---` fence** — and 12 markdown sections (2 conditional). Copy the `## Template` block below verbatim and fill the bracketed placeholders. **Domain enum** (`frontmatter.domain`, required): | Deliverable signature | domain value | |---|---| | Activity / exhibition / event signage | `poster` | | Standalone illustrative artwork, no host | `illustration` | | Product packaging / bottle / label | `packaging` | | Brand visual system / marketing series / peripherals | `brand_visual` | | Book / magazine / report cover | `editorial_cover` | | Brief for an upcoming photo shoot | `photography_brief` | | Hero illustration / splash art for app or web (not layout) | `hero_visual_for_ui` | **Disambiguation**: art-exhibition poster → `poster` (activity-signage wins over editorial); book-with-illustrated-cover → `editorial_cover` (host-artifact wins over illustration). ### Template (copy and fill) ````markdown --- slug: YYYY-MM-DD- status: draft domain: tradition: # NEVER "N/A", "none", "", "unknown" — see B7 style_treatment: # from Style treatment question generated_by: visual-brainstorm@0.1.0 created: YYYY-MM-DD updated: YYYY-MM-DD --- # > Partial OpenSpec alignment: status enum + RFC 2119 keywords. Does not use ADDED/MODIFIED/REMOVED delta markers. Full alignment deferred until /visual-spec consumption validates need. ## Intent <2-5 sentences of user intent; no compositional detail> ## Audience ## Physical form ## Market <国内 / 海外 / 多语言; or "domestic, no multilingual"> ## Budget & deadline