Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Guidance for creating and editing designs using the designer tool.
tools
["designer"]
tags
[]
DESIGNER TOOL:
You have designer tools that create/edit multi-page visual designs.
Each project has a mode, a canvas aspect ratio, multi-page content, and an optional brand config.
All designs are rendered as HTML/CSS in a sandboxed iframe preview.
Project modes (pick one — the mode dictates layout, canvas, and tool surface)
mode
What it produces
page_kind
Default aspect
Canvas behaviour
deck
Slide deck (16:9 / 4:3 / 1:1 / 9:16)
slide
16:9 (1920×1080)
Fixed body — width:Wpx; height:Hpx; overflow:hidden. Each page is a single slide.
document
A4 / Letter document pages
slide
A4 (794×1123)
Fixed page body — same fixed-frame rules as deck.
landing
Tall scrollable landing page
screen
landing (1440×3200 hint)
Tall scrollable. Use html,body{min-height:100vh;overflow-x:hidden} and a .page{max-width:1440px;margin:0 auto} wrapper. NEVER set body{overflow:hidden} or fixed pixel height. Use responsive units (vw, clamp(), %).
app_mockup
Phone or desktop UI prototype
screen
phone (390×844)
One device viewport per page/route. html,body{width:Wpx;height:Hpx;overflow:hidden}. In-content scrolling goes inside an inner .screen-body{overflow-y:auto}.
storyboard
Cinematic shot-by-shot board
shot
16:9 (1920×1080)
Fixed-frame like deck. Each page = one shot with beat / caption / camera direction.
The user picks mode in the New Design dialog. You can read it with designer_get_project and change it with designer_set_mode. Switching modes does NOT auto-resize the canvas — call designer_resize_project after designer_set_mode if you also want a different aspect.
Core page operations
designer_set_pages: Replace ALL pages. Use for new projects or full reworks. Input: list of {html, title, notes}.
designer_update_page: Update a single page HTML. Input: index, html, optional title.
designer_add_page: Insert a new page. Use index=-1 to append.
designer_delete_page: Remove a page by index. DESTRUCTIVE — requires user approval.
designer_move_page: Reorder a page. Input: from_index, to_index.
designer_get_project: Read project summary (mode, page titles, brand, dimensions). Call before any multi-page edit.
designer_get_page_html: Read the full stored HTML for a single page. Call before full-page rewrites so you preserve existing assets and layout.
designer_get_reference: Read a saved project reference by id or filename so you can reuse prior attachments without asking again.
Interactive modes — landing and app_mockup
For these modes, never write raw <script> or onclick=. The preview is sandboxed and a thin runtime bridge wires interactivity from declarative attributes.
Declarative attributes the runtime understands
data-row-bot-route="<route_id>" → marks a screen/route section. The runtime shows one route at a time. The outer route container also gets data-row-bot-route-host="1" automatically when emitted via designer_add_screen.
data-row-bot-action="navigate:<route_id>" → click handler that switches to the named route.
data-row-bot-action="toggle_state:<key>" → click handler that flips a state key. The runtime sets data-row-bot-state-<key>="on|off" on <html>; style with [data-row-bot-state-<key>="on"] selectors.
data-row-bot-action="play_media:<asset_id>" → click handler that plays a video/audio asset.
data-row-bot-transition="fade|slide_left|slide_up|none" (optional, on the navigation source).
Tools for interactive projects
designer_add_screen: Add a new screen/route. Input: title, optional route_id (auto-slugified from title), optional html (branded blank if empty), optional copy_from (page index to duplicate).
designer_link_screens: Wire a click on one element to navigate to another screen. Input: source_route, selector (CSS or data-row-bot-element-id), target_route, optional transition.
designer_reorder_routes: Reorder the route list (which is the page list for screen-mode projects).
designer_preview_screen: Render a single route in isolation for review.
designer_set_mode: Switch the project's mode (and tool surface).
Authoring rules per interactive mode
landing: usually ONE page that scrolls vertically. Sections (<section id="...">) for hero / features / pricing / FAQ / CTA / footer. Use anchor links (<a href="#features">) for in-page jumps; reserve data-row-bot-action="navigate:..." for cross-route flows in multi-page landings.
app_mockup: ONE route per logical screen (Home, Detail, Settings, etc.). Each page's HTML is the entire device viewport. Use data-row-bot-action="navigate:<route_id>" on tappable rows, tab bars, and back buttons. Use data-row-bot-transition="slide_left" when going deeper, slide_right when going back. Wrap scrollable content inside an inner .screen-body{overflow-y:auto}.
Media (images, video)
designer_generate_image: Generate an AI image from a text prompt and embed it in a page. Input: prompt, optional page_index (-1=active), position (top/bottom), width, height, size. When a page contains a data-row-bot-shot-visual or data-row-bot-image-slot placeholder (storyboards, typed deck slots), the generated image automatically replaces the placeholder and fills the slot — no extra positioning call needed for the first generation.
designer_insert_image: Insert an attached or local image file. Same slot-replacement behavior as generate.
designer_generate_video: Generate an MP4 from a text prompt (or image-to-video). Input: prompt, optional page_index, position, width, aspect_ratio. Embedded as <video autoplay loop muted playsinline> with a poster.
designer_insert_video: Insert an attached / local video file (mp4/webm/mov). Same autoplay/loop/muted defaults.
designer_remove_image: Remove an inserted image, chart, or video from a page WITHOUT deleting the page. Input: image_ref (asset ID or label), optional page_index (-1=active). Use this — NOT designer_delete_page and NOT designer_replace_image — when the user says "remove the picture", "delete the image from shot 2", "clear the visual", etc. Shot-visual / image-slot placeholders revert to their dashed preview automatically.
When reusing a known project asset inside hand-written HTML, use src="asset://ASSET_ID" and keep data-asset-id="ASSET_ID" on the <img> / <video>. Never invent placeholder tokens like __ASSET_...__ or __REMOVE__.
AI content & polish
designer_refine_text: Refine a text element. Input: page_index, tag (e.g. 'h1', 'p'), old_text (exact text), action (shorten/expand/professional/casual/persuasive/simplify/bullets/paragraph/custom), optional custom_instruction.
designer_add_chart: Add a chart. Input: chart_type (bar/line/pie/scatter/donut/histogram/box/area/heatmap), data_csv (inline CSV with header), optional title, page_index, position.
designer_critique_page: Review for hierarchy, overflow, contrast, readability, spacing.
designer_apply_repairs: Apply safe deterministic fixes for selected critique categories.
designer_brand_lint: Read-only scan for contrast issues, off-palette colors, non-brand fonts, missing alt text, logo safe-zone overlaps. Input: optional page_index (-1=all). Returns structured JSON.
designer_generate_notes: Generate speaker notes for one page (deck/document modes).
Brand & canvas
designer_set_brand: Update brand colors/fonts and logo placement. Includes logo_mode, logo_scope, logo_position, logo_max_height, logo_padding.
designer_resize_project: Resize the canvas using a built-in preset or explicit aspect ratio. Available aspects: 16:9, 4:3, 1:1, 9:16, A4, letter, landing (1440×3200), phone (390×844), desktop (1440×900). Use after designer_set_mode if changing both.
Export & publish
designer_export: Input: format (pdf/html/png/pptx), optional page range. PDF/PNG/PPTX work best for deck and document. Single-file html works for any mode.
designer_publish_link: Publish a self-contained shareable HTML deck link. For landing and app_mockup, this produces an interactive bundle (includes the runtime bridge + transitions CSS) so navigation and toggles work in the published page.
HTML rules (all modes)
Each page MUST be a complete self-contained HTML document with inline <style>.
Use CSS variables for brand: --primary, --secondary, --accent, --bg, --text, --heading-font, --body-font.
Canvas size + mode are provided in the system prompt — follow the canvas rules table above.
Include Google Fonts <link> if using custom fonts.
Modern CSS only: flexbox, grid, gradients, box-shadow. No external frameworks.
For placeholder images: colored SVG shapes or gradient divs, never external URLs.
No <script> and no inline onclick — the iframe is sandboxed. Use data-row-bot-action for interactive modes; deck/document/storyboard pages have no interactivity.
Content budgets per mode (anti-clipping)
Fixed-slide modes (deck, document, storyboard) clip with overflow:hidden — anything past the canvas height is lost. Stick to these per-page budgets:
deck (16:9 @ 1920×1080): ONE heading + EITHER (a) one paragraph ≤45 words, (b) up to 3 cards/columns with ≤25 words each, OR (c) one chart/image + ≤2 bullets. Max 5 bullets. 64–96 px edge padding. Heading ≤4.5rem, body ≤1.4rem, line-height 1.3–1.5. Leave ≥48 px of bottom breathing room.
document (A4 / Letter): 130–160 words body copy per page (lead paragraph + 3–4 short sections with ~2 bullets each). Max ~5 top-level sections per page — split into more pages rather than cramming. Generous line-height (1.45–1.6), 16–24 px gap between sections, ≥32–48 px bottom padding.
storyboard (1920×1080): each shot gets ONE eyebrow (≤3 words) + ONE heading (≤6 words) + ONE short paragraph (≤35 words) OR a pull-quote card + AT MOST two small metadata cards (shot type, camera move) OR one voiceover card — not both + ONE direction/note line (≤30 words) + ONE footer strip (timeline OR shot counter). Pick 3–4 visible blocks max. Push extra detail to page notes via designer_generate_notes instead of the visible frame. Right column uses display:flex; flex-direction:column; gap:20–28px; padding:64px 56px; with ≥48 px bottom space.
Landing and app_mockup modes flow differently — landing pages scroll vertically, app mockups fit one device viewport per route — but the same principle applies: count blocks before you write HTML, and split rather than cram.
Authoring guardrails (all modes)
No decorative overlap on text. Never place absolutely-positioned decorative CSS art (blobs, hand-drawn shapes, mascots, chef figures, etc.) on top of a heading or body paragraph. If a hero has a headline, either (a) give the illustration its own column/row, (b) put the illustration behind the text with low opacity and z-index:0 AND ensure the text has a readable background or text-shadow, or (c) omit the illustration. When in doubt, use a real AI image via a typed image slot instead of CSS shapes.
Button rows stay horizontal. When emitting two buttons side-by-side (Back + primary, Cancel + Confirm, etc.), use display:flex; gap:12–24px; and give the primary button flex:1. Never let them stack vertically or touch. Secondary / ghost buttons must be visually distinct (transparent or outlined) from the primary action — never two identical filled pills.
Typed image slots over overlays. When authoring a page that will later receive an AI image, mark the target container with data-row-bot-image-slot="NAME" (e.g. <div data-row-bot-image-slot="hero" style="width:100%;aspect-ratio:16/9;"></div>). designer_generate_image / designer_insert_image will fill the first matching slot, sized to cover. Alternatively pass position="replace:.my-class" or position="replace:#my-id" to target a specific container. Never emit an absolute-positioned image overlay unless the user explicitly asked for a floating element.
Post-critique repair loop (mandatory after major rewrites)
After any full-page rewrite, full-project rebuild, mode switch, canvas resize, or multi-page update:
Call designer_critique_page(page_index=<n>) on each changed page.
If any finding category is overflow, either:
call designer_apply_repairs(page_index=<n>, categories=["overflow"]), OR
trim content and call designer_update_page again, OR
split to a new page via designer_add_page.
Never ship a page with an unresolved overflow finding.
For broader sweeps, pass multiple categories: designer_apply_repairs(categories=["spacing","readability","overflow"]).
Workflow patterns
Decks / documents / storyboards (static)
Creating: designer_get_project → designer_set_pages with all pages.
Editing one slide: designer_update_page with the affected index.
"Make all pages darker": designer_get_project → designer_update_page per page.
"Add a pricing slide after slide 3": designer_add_page(index=3, ...).
"Export as PDF": designer_export(format="pdf").
Speaker notes: designer_generate_notes for the relevant page (don't rewrite the HTML).
Landing pages
One page is fine. Build vertically with <section> blocks: hero, features, pricing, FAQ, CTA, footer.
Use clamp() and vw units for typography. Use max-width:1440px on a .page wrapper.
For multi-page landing flows (e.g. a stepper), use designer_add_screen per step and designer_link_screens for the navigation.
App mockups
Plan routes first: list every screen the user mentions. Call designer_set_pages (or designer_add_screen per route) so each page has data-row-bot-route="<id>" on its outer container.
Wire navigation explicitly with designer_link_screens or designer_set_interaction(action="navigate", ...). Don't use <a href="#detail">.
For toggleable UI (dark mode, notifications, expanded panels), use data-row-bot-action="toggle_state:<key>" and style with [data-row-bot-state-<key>="on"] selectors on <html>.
Wrap scrollable content inside an inner .screen-body — the outer body stays at the device viewport.
Mode change
"Turn this deck into a landing page": designer_set_mode("landing") → designer_resize_project("landing") → designer_set_pages([...]) rewriting the content as a single tall page.
Canvas / brand
Canvas changes: designer_resize_project before restyling when the user wants square, vertical, A4, Letter, phone, desktop, landing, or standard slide formats.
Brand changes: designer_set_brand first. It updates stored brand CSS and can switch the automatic logo overlay between all pages, first page only, or manual placeholder mode.
Sharing
Shareable deck/landing URL: designer_publish_link (handles interactive bundles automatically when mode is landing or app_mockup).
One-off file: designer_export(format="html") for a single self-contained file.
AI imagery
"Add a photo of mountains": designer_generate_image(prompt="cinematic photo of mountain landscape at sunrise") — or ask the user to attach a reference image and call designer_insert_image.
"Generate an AI image of a futuristic city": designer_generate_image(prompt="futuristic city skyline").
"Add a 4-second product video": designer_generate_video(prompt="...", aspect_ratio="16:9").
"Generate a still image for every shot": call designer_generate_image once per page — shot-visual placeholders are filled automatically in the correct slot for each.
"Remove the picture from shot 2" / "delete the image on this page": call designer_remove_image(image_ref="<asset-id or label>", page_index=<n>). Do NOT use designer_delete_page (that removes the whole page) and do NOT pass a fake sentinel to designer_replace_image.
Components & critiques
"Add a metrics strip near the top": designer_insert_component(component_name="stats_band", page_index=-1, position="top").
"Review the current slide and fix what feels cramped": designer_critique_page(page_index=-1) → designer_apply_repairs(page_index=-1, categories=["spacing","readability","overflow"]).
"Make the heading shorter": designer_refine_text(page_index=0, tag="h1", old_text="...", action="shorten").
"Add a bar chart of Q1 revenue": designer_add_chart(chart_type="bar", data_csv="Quarter,Revenue\nQ1,120\nQ2,150\n...").
"Show me what you changed" / "review last turn": the UI surfaces a review dialog with a page-by-page mutation diff after every agent turn — you don't need to call it, but do reference it when the user asks what you altered.
IMPORTANT
Refs from designer_get_project can go stale. Re-read if unsure.
Always explain what you changed in your text response after tool calls.
Maintain visual consistency across pages.
Keep HTML compact — avoid unnecessary nesting or unused CSS.
AI images: descriptive prompts. Specify style (photo, illustration, etc.). There is no stock-image tool — use designer_generate_image or designer_insert_image (attached/local file) only.
For interactive modes, never emit <script>, inline event handlers, or rely on <a href> for navigation between routes — use data-row-bot-action.
