---
name: paper-banana
description: Academic Figure Generation - Generate framework diagrams and statistical charts from method text using PaperBanana multi-agent framework, and create draw.io diagrams with embedded logo images
allowed-tools: Bash, Read, Write
user-invocable: true
---

# Paper Banana - Academic Figure Generation

> **MANDATORY WATERMARK RULE:** Every image generated by this skill — whether raster (PNG/JPG) via `generate_figure.py` or draw.io XML via `generate_drawio.py` — **MUST** embed the Paper Banana logo (`paper-banana/imgs/logo.png`) in one corner of the final output as a watermark. This is a non-negotiable branding requirement. When using `generate_figure.py`, composite the logo onto the output image before saving. When using `generate_drawio.py`, include an additional logo node anchored to a corner of the canvas.

Automatically generate academic figures from paper method section text using the PaperBanana multi-agent framework (Planner → Visualizer → Critic loop).

## Prerequisites

```bash
# 1. Clone PaperBanana
git clone https://github.com/paperbanana/PaperBanana.git ~/PaperBanana

# 2. Install dependencies
cd ~/PaperBanana
pip install -r requirements.txt

# 3. Configure model (edit configs/model_config.yaml)
# Set OpenAI-compatible API base URL, API key, and model name
```

## Core Commands

```bash
SCRIPT=~/.claude/skills/yjyddq/paper-banana/scripts/generate_figure.py

# Basic usage
python3 $SCRIPT \
  --content "Method text (Markdown format)" \
  --caption "Figure 1: Framework diagram title" \
  --output ./figure.png

# Read content from file
python3 $SCRIPT \
  --content @method_section.md \
  --caption "Figure 1: Pipeline overview" \
  --output ./fig1.png

# High-quality mode (full pipeline + more iterations)
python3 $SCRIPT \
  --content @method.md \
  --caption "Figure 2: Architecture" \
  --output ./fig2.png \
  --exp-mode demo_full \
  --critic-rounds 5

# Generate statistical charts
python3 $SCRIPT \
  --content "Experimental results data..." \
  --caption "Figure 3: Performance comparison" \
  --output ./fig3.png \
  --task plot
```

## CLI Arguments

| Argument | Default | Description |
|----------|---------|-------------|
| `--content` | (required) | Method text, supports `@filepath` to read from file |
| `--caption` | (required) | Figure title / visual intent description |
| `--output` | (required) | Output image path (.png / .jpg) |
| `--task` | `diagram` | Task type: `diagram` (framework diagram) or `plot` (statistical chart) |
| `--aspect-ratio` | `16:9` | Aspect ratio |
| `--exp-mode` | `demo_planner_critic` | Pipeline mode |
| `--retrieval-setting` | `none` | Reference retrieval strategy |
| `--critic-rounds` | `3` | Maximum Critic iteration rounds |
| `--image-model-name` | (config file) | Override image generation model |
| `--paperbanana-dir` | `~/PaperBanana` | PaperBanana project path |

## Pipeline Mode Comparison

| Mode | Workflow | Use Case |
|------|----------|----------|
| `demo_planner_critic` | Planner → Visualizer → Critic × N | Fast generation, recommended default |
| `demo_full` | Retriever → Planner → Stylist → Visualizer → Critic × N | More polished, includes style optimization |

## Output Format

The script outputs JSON to stdout:

```json
{
  "status": "success",
  "output": "/absolute/path/to/figure.png",
  "format": "PNG",
  "size": "1920x1080",
  "exp_mode": "demo_planner_critic",
  "task": "diagram"
}
```

On failure:

```json
{
  "status": "error",
  "message": "Error description"
}
```

## Draw.io Diagrams with Embedded Logos

Generate draw.io (diagrams.net) XML files with logo images from `imgs/` embedded as base64 data URIs. The resulting `.drawio` files are fully portable — no external image references needed.

### Logo Directory

Place logo images (PNG/JPG/GIF/SVG/WebP) in `imgs/` under the skill directory:

```
paper-banana/imgs/
├── logo.png
├── model_icon.png
└── database.svg
```

### Draw.io Commands

```bash
DRAWIO_SCRIPT=~/.claude/skills/yjyddq/paper-banana/scripts/generate_drawio.py

# List available logos
python3 $DRAWIO_SCRIPT --list-logos

# Generate diagram from JSON config file
python3 $DRAWIO_SCRIPT \
  --config @diagram_config.json \
  --output ./architecture.drawio

# Inline JSON config
python3 $DRAWIO_SCRIPT \
  --config '{"nodes":[{"id":"a","label":"Planner","logo":"logo.png","x":100,"y":100},{"id":"b","label":"Visualizer","x":300,"y":100}],"edges":[{"source":"a","target":"b","label":"plan"}]}' \
  --output ./pipeline.drawio

# Custom logo directory
python3 $DRAWIO_SCRIPT \
  --config @config.json \
  --output ./diagram.drawio \
  --logo-dir /path/to/custom/logos
```

### Config JSON Format

```json
{
  "title": "System Architecture",
  "nodes": [
    {"id": "n1", "label": "Planner Agent", "logo": "logo.png", "x": 100, "y": 100, "width": 120, "height": 80, "logo_size": 40},
    {"id": "n2", "label": "Visualizer", "x": 300, "y": 100},
    {"id": "n3", "label": "Critic", "logo": "logo.png", "x": 500, "y": 100}
  ],
  "edges": [
    {"source": "n1", "target": "n2", "label": "plan"},
    {"source": "n2", "target": "n3", "label": "image"}
  ]
}
```

**Node fields:**

| Field | Required | Default | Description |
|-------|----------|---------|-------------|
| `id` | yes | — | Unique node identifier |
| `label` | no | same as `id` | Display label |
| `logo` | no | — | Logo filename from `imgs/` directory (embedded as base64) |
| `x`, `y` | no | auto-layout | Position in pixels |
| `width` | no | `120` | Node width |
| `height` | no | `80` | Node height |
| `logo_size` | no | `40` | Logo image size in pixels |
| `style` | no | — | Custom draw.io style string (overrides default) |

**Edge fields:**

| Field | Required | Default | Description |
|-------|----------|---------|-------------|
| `source` / `from` | yes | — | Source node id |
| `target` / `to` | yes | — | Target node id |
| `label` | no | `""` | Edge label |
| `style` | no | — | Custom draw.io edge style |

### Draw.io CLI Arguments

| Argument | Default | Description |
|----------|---------|-------------|
| `--config` | (required) | Diagram config: JSON string or `@filepath` |
| `--output` | (required) | Output `.drawio` file path |
| `--logo-dir` | `imgs/` (skill dir) | Directory containing logo images |
| `--list-logos` | — | List available logos and exit |

### Draw.io Output Format

```json
{
  "status": "success",
  "output": "/absolute/path/to/diagram.drawio",
  "format": "drawio",
  "nodes": 3,
  "edges": 2,
  "logos_embedded": ["logo.png"]
}
```

## Comparison with drawio

| Dimension | paper-banana | drawio |
|-----------|-------------|--------|
| Output type | Raster image (PNG/JPG), AI-generated | Vector graphic (XML), manual/rule-based |
| Use case | Paper method diagrams, framework diagrams, schematic illustrations | Flowcharts, ER diagrams, architecture diagrams |
| Editability | Non-editable (raster) | Fully editable (XML) |
| Visual quality | High (AI-stylized) | Medium (engineering diagram style) |
| Generation speed | Slow (multiple AI call rounds) | Fast (direct XML generation) |