Add creative design skills (landing-page-design, tastemaker, emil-design-eng, animate, better-colors, better-layout, better-typography, web-design-engineer) and update design-build & design-explore

This commit is contained in:
2026-09-18 13:57:37 +10:00
parent c37e28149f
commit 6c3d11aa73
10 changed files with 2459 additions and 164 deletions

View File

@@ -1,11 +1,16 @@
---
name: design-explore
description: Generate UI design mockups as images for exploration and ideation. Use when the user wants to visualize a design concept before building. Supports optional DESIGN.md from awesome-design-md for design-token-accurate mockups.
description: Generate UI design mockups as images for exploration and ideation. Uses Tastemaker aesthetics, landing-page structure, and DESIGN.md tokens to visualize creative, non-generic UI directions before building.
---
# /design-explore
Generate visual design mockups. You ask the user what they want to design, optionally pull design tokens from a DESIGN.md file, and call the image-maker sub-agent to produce polished images.
Generate visual design mockups. Asks the user what they want to design, optionally pulls design tokens from a DESIGN.md file, and calls the `image-maker` sub-agent to produce polished mockup images.
Utilizes guidance from installed design skills:
- **`tastemaker`**: Avoid generic AI slop (indigo-to-purple gradients, generic soft cards). Express distinct aesthetic taste.
- **`landing-page-design`**: Structured hero sections, headline/CTA hierarchy, conversion layouts.
- **`web-design-engineer`**: High-craft visual presentation and browser acceptance standards.
## Flow
@@ -15,57 +20,42 @@ Ask the user two questions:
1. **"What are we designing?"** — Free text description of the interface, page, or component.
2. **"What style? Pick from awesome-design-md or describe it."** — The user can either:
- Provide a full URL: `https://getdesign.md/apple/design-md` or from `https://github.com/VoltAgent/awesome-design-md`
- Describe the style in words: "dark glass morphism, purple accent, rounded corners"
- If they provide a URL, download it and extract the design tokens
2. **"What style or vibe? Pick from awesome-design-md, describe it, or reference a look."**
- Provide a full URL (e.g. `https://github.com/VoltAgent/awesome-design-md`)
- Describe in words: "warm editorial, high-contrast dark, Swiss typography, neo-brutalist"
### Step 2: If DESIGN.md URL provided, extract tokens
### Step 2: Extract tokens & design guidelines
Download the DESIGN.md file:
If a DESIGN.md URL is provided, download it:
```bash
curl -sL <DESIGN_MD_URL> -o /tmp/design-tokens.md
```
Read the file and extract these tokens from the YAML/Markdown:
- **Palette/Colors** — background, primary, accent, text, surface
- **Typography** — font family, heading font, sizes, weights
- **Spacing** — base unit, section padding
- **Border Radius** — button, card, input values
- **Shadows** — elevation levels
- **Identity/Language** — any prose describing the visual language
If the file is YAML, parse it with `python3 -c "import yaml; ..."`. If it's Markdown, read the relevant sections.
Extract palette, typography, spacing, border radius, and identity notes.
### Step 3: Build the image prompt
Combine the user's design description with the extracted tokens into a detailed prompt:
Combine description and tokens into a prompt for `image-maker`:
```
Design prompt for {what}: {user's description}
UI mockup for {what}: {user's description}
Design system tokens to follow exactly:
Design system tokens to follow:
- Color palette: {extracted palette}
- Typography: {extracted fonts}
- Border radius: {extracted radius values}
- Spacing: {extracted spacing}
- Visual identity: {extracted identity description}
- Radius & Spacing: {extracted radius/spacing}
- Visual Vibe: {extracted identity description}
Style requirements:
- Professional, high-fidelity UI mockup
- Show the interface in context (browser frame or device mockup)
- Clean, modern presentation
- Use the exact colors and fonts specified
Aesthetic requirements (from Tastemaker & Web Design Engineer):
- High-fidelity, stunning UI mockup in a clean browser frame or device context
- NO generic AI slop: avoid default indigo-to-purple gradients, generic soft cards, or template layouts
- Distinct visual identity with strong typographic scale and intentional color hierarchy
- Clean alignment, generous whitespace, and sharp contrast
```
If NO DESIGN.md, use only the user's description with general style guidance.
### Step 4: Call image-maker
Launch the image-maker sub-agent with the prompt:
```bash
Agent(
subagent_type: "image-maker",
@@ -75,74 +65,13 @@ Agent(
)
```
The image-maker will generate images via OpenRouter's image models.
### Step 5: Show images inline (REQUIRED)
After images are generated, you MUST display them inline. Do NOT just save files and report paths.
1. Save each image to `/tmp/design-explore-<N>.png`
2. **Call `read` on each saved image file** — pi displays images as inline attachments
After images are generated:
1. Save images to `/tmp/design-explore-<N>.png`
2. Call `read` on each saved image file so Pi displays them inline in the conversation.
3. Present to user: "Here are the design concepts. Which direction?"
Example:
```bash
# After image-maker generates images, save them
# Then READ them so they appear inline:
read("/tmp/design-explore-1.png")
read("/tmp/design-explore-2.png")
read("/tmp/design-explore-3.png")
```
### Step 6: Iterate or Build
The user should see the images in the conversation, not have to open files manually.
### Step 6: Iterate
The user can:
- Pick a direction: "Go with #2 but change the accent to teal"
- Refine: generate updated images with the feedback
- Move on: "I like this. Let's build it with design-build"
---
## DESIGN.md File Format
DESIGN.md files typically look like:
```yaml
---
palette:
background: "#FFFFFF"
primary: "#007AFF"
accent: "#FF9500"
text: "#1D1D1F"
typography:
heading: "SF Pro Display"
body: "SF Pro Text"
sizes:
h1: "48px"
h2: "32px"
body: "17px"
spacing:
unit: "8px"
section: "80px"
radius:
button: "12px"
card: "16px"
input: "8px"
shadows:
card: "0 2px 8px rgba(0,0,0,0.08)"
elevated: "0 8px 24px rgba(0,0,0,0.12)"
identity: Clean, minimal, generous white space, Apple-like clarity
---
```
But the exact format varies — always read the file contents to know the structure.
---
## Resources
- **awesome-design-md**: https://github.com/VoltAgent/awesome-design-md (55+ design systems)
- **getdesign.md**: https://getdesign.md/ (hosted DESIGN.md files)
- **image-maker agent**: Uses OpenRouter's gpt-5-image-mini (~4¢/image)
User can pick a direction, request tweaks, or run `/design-build` to generate the HTML/Tailwind implementation.