5,366,387 AI resources indexedWorking toward the world’s largest AI resource index.

PDF

images

Generate blog post thumbnails for Skill Stack using the brand aesthetic. Follows an iterative workflow - brainstorm concepts, get approval, generate with Gemini API.

Installable skillMITBy cdeistopenedUpdated Oct 27, 2026
INTENDED FOROpenClawInstall in a listed agent.

instructions.md

---
name: skill-stack-thumbnails
description: Generate blog post thumbnails for Skill Stack using the brand aesthetic. Follows an iterative workflow - brainstorm concepts, get approval, generate with Gemini API.
---

# Skill Stack Thumbnail Generator

Generate on-brand thumbnails for Skill Stack blog posts using Google's Gemini API.

## Brand Identity

**Skill Stack** is a curated repository of SKILL.md files for content creators. The aesthetic is:
- **Clean and minimal** - Anthropic-adjacent sophistication
- **Paper/document metaphor** - Skills as stackable knowledge artifacts
- **Terracotta accent** - Single warm pop of color
- **Editorial quality** - Think publishing meets productivity

### Logo Reference

The logo is 4 stacked paper documents:
- 3 gray layers (light → medium → darker)
- 1 terracotta top layer with folded corner
- Warm cream background
- Subtle shadows for depth

## Color Palette

| Role | Color | Hex |
|------|-------|-----|
| **Primary Accent** | Terracotta/Coral | `#D4654A` |
| **Background** | Warm Cream | `#F5F3EE` |
| **Paper Light** | Light Gray | `#E8E6E1` |
| **Paper Medium** | Medium Gray | `#DEDBD5` |
| **Paper Dark** | Darker Gray | `#D4D1CB` |
| **Text/Line** | Near-Black | `#2D2D2D` |

**Rule**: Terracotta appears on ONE element only per image.

---

## Workflow

### Step 1: Understand the Post

Read the post title and content to understand:
- What's the core concept?
- What metaphor would resonate?
- What visual would make someone click?

### Step 2: Brainstorm Concepts (APPROVAL REQUIRED)

Generate 4-5 visual concepts. Each should be:
- **One sentence** describing the visual
- **Concrete** - you can picture it instantly
- **Non-generic** - no lightbulbs, handshakes, arrows
- **On-brand** - fits the paper/document aesthetic

**Format for presenting concepts:**

```
## Thumbnail Concepts for "[Post Title]"

1. **[Short label]** - One sentence visual description. Why it works.

2. **[Short label]** - Description...

3. **[Short label]** - Description...

4. **[Short label]** - Description...

Which direction resonates? I can refine from there.
```

**Wait for user approval before proceeding.**

### Step 3: Develop the Prompt

Once user selects a concept, expand into full prompt:

```
Create a minimal editorial thumbnail for a blog post about [TOPIC].

CONCEPT: [Expand the selected concept into 2-3 sentences]

STYLE: Clean, flat editorial illustration with paper-like shapes and subtle shadows.
Think Anthropic brand guidelines meets publishing design - sophisticated, minimal, intentional.
NOT photorealistic. NOT glossy AI aesthetic. NOT busy.

COMPOSITION:
- Focal element offset to [left/right third]
- Generous negative space (40%+ of frame empty)
- Single clear focal point
- Asymmetrical balance

COLOR PALETTE:
- Background: warm cream, almost off-white (like #F5F3EE)
- Primary accent: terracotta/burnt coral (like #D4654A) on ONE key element only
- Supporting: light gray and medium gray tones for layered paper shapes
- Shadows: soft, subtle, not harsh drop shadows

TEXTURE: Matte paper quality. Flat shapes with slight edge shadows.
No glossy renders. No gradients. No photorealism.

AVOID:
- Busy compositions
- More than 2-3 distinct elements
- Photorealistic rendering
- Cliche imagery (lightbulbs, gears, arrows)
- Neon or saturated colors
- Gradients
- Glossy AI aesthetic
- Text or words in the image

FORMAT: 16:9 landscape aspect ratio for blog thumbnails
```

### Step 4: Generate

Run via Gemini API:

```bash
cd scripts/images
export GEMINI_API_KEY=$(grep GEMINI_API_KEY ../../.env.local | cut -d= -f2)
python3 generate_image.py "PROMPT" --output ../../public/images/thumbnails --name "post-slug"
```

Or use the `/gemini-imagegen` skill if available.

### Step 5: Review & Iterate

After generation:
- **80% good?** Request specific edits
- **Composition off?** Adjust framing language
- **Wrong vibe?** Simplify further or try different concept
- **Too busy?** Remove elements

---

## Concept Templates by Post Type

### Framework Posts (CODER, 4S, etc.)
- Stacked/layered shapes representing steps
- Flow diagrams with paper elements
- Numbered paper cards in sequence

### Philosophy/Thesis Posts
- Transformation metaphors (rough → polished)
- Contrast/juxtaposition of two elements
- Single powerful symbol

### Tool/Technical Posts
- Tool icon + paper document interaction
- Simple geometric representation
- Integration/connection visual

### How-To/Practical Posts
- Action in progress (folding, stacking, organizing)
- Before/after implied through arrangement
- Single focused action

---

## Example Concepts

**For "The 4S Framework":**
1. **Four stacked squares** - Four paper squares arranged diagonally ascending, each labeled S1-S4, top one terracotta
2. **Funnel of papers** - Four documents funneling into one polished output
3. **Compass with 4 points** - Vintage compass where N/E/S/W are replaced by the 4 S's
4. **Building blocks** - Four cube shapes stacking, architectural quality

**For "Transform Don't Generate":**
1. **Crumpled to crisp** - Crumpled paper ball on left, crisp folded document on right, terracotta accent on the crisp one
2. **Rough cut to gem** - Raw stone transforming into faceted shape
3. **Sketch to blueprint** - Messy sketch evolving into clean technical drawing

---

## Output Location

Save thumbnails to: `public/images/thumbnails/[post-slug].png`

Reference in frontmatter: `image: "/images/thumbnails/[post-slug].png"`

---

*Always get concept approval before generating. The thinking is as important as the image.*

Discuss this resource

Platform guidance

Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.

Resource notes updated 2026-10-27 · Original instructions by cdeistopened