sf-imagen: Salesforce Visual AI Skill Skill

sf-imagen: Salesforce Visual AI Skill

Visual content generation and AI sub-agent for Salesforce development using Gemini CLI with Nano Banana Pro extension.

⚠️ IMPORTANT: Prerequisites Check

Before using this skill, ALWAYS run the prerequisites check first:

~/.claude/plugins/marketplaces/sf-skills/sf-imagen/scripts/check-prerequisites.sh

If the check fails, DO NOT invoke this skill. The user must fix the missing prerequisites first.

Requirements

| Requirement | Description | How to Get | |-------------|-------------|------------| | macOS | Required for Preview app image display | Built-in | | GEMINI_API_KEY | Personal API key for Nano Banana Pro | https://aistudio.google.com/apikey | | Gemini CLI | Command-line interface for Gemini | npm install -g @google/gemini-cli | | Nano Banana Extension | Image generation extension | gemini extensions install nanobanana |

Optional (for 4K/editing via Python script)

| Requirement | Description | How to Get | |-------------|-------------|------------| | uv | Fast Python package runner | curl -LsSf https://astral.sh/uv/install.sh \| sh |

Setting Up API Key

Add to ~/.zshrc (DO NOT commit to version control):

export GEMINI_API_KEY="your-personal-api-key"
export NANOBANANA_MODEL=gemini-3-pro-image-preview

Core Capabilities

1. Visual ERD Generation

Generate actual rendered ERD diagrams (not just Mermaid code):

Query object metadata via sf-metadata
Generate visual diagram with Nano Banana Pro
Display in macOS Preview app (default)

2. LWC/UI Mockups

Generate wireframes and component mockups:

Data tables, record forms, dashboard cards
Experience Cloud page layouts
Mobile-responsive designs following SLDS

3. Gemini Code Review (Sub-Agent)

Parallel code review while Claude continues working:

Apex class/trigger review for best practices
LWC component review for accessibility
SOQL query optimization suggestions

4. Documentation Research (Sub-Agent)

Parallel Salesforce documentation research:

Look up API references and limits
Find best practices and patterns
Research release notes

🔄 Draft → Final Workflow (Cost-Effective Iteration)

For complex visuals, iterate at low resolution before generating the final:

| Phase | Resolution | Purpose | Time | Method | |-------|------------|---------|------|--------| | Draft | 1K | Quick feedback, prompt refinement | ~3s | CLI or Python | | Iteration | 1K | Style/layout adjustments | ~3s | CLI /edit or Python | | Final | 4K | Production-quality output | ~10s | Python script only |

Recommended Workflow

Draft at 1K - Generate initial image quickly
Iterate - Refine prompt or use /edit to adjust
Final at 4K - Once satisfied, generate production quality

Example: Draft to Final

# Step 1: Draft (fast, cheap)
gemini --yolo "/generate 'Account-Contact ERD with blue boxes'"
open ~/nanobanana-output/*.png  # Review

# Step 2: Iterate with /edit
gemini --yolo "/edit 'Add Opportunity, use thicker arrows'"

# Step 3: Final at 4K (requires Python script)
uv run scripts/generate_image.py \
  -p "Account-Contact-Opportunity ERD, blue boxes, thick arrows, legend" \
  -f "crm-erd-final.png" \
  -r 4K
open ~/nanobanana-output/*crm-erd-final*.png

Image Editing with /edit

Refine existing images with natural language instructions:

# Edit the most recent generated image
gemini --yolo "/edit 'Move Account box to center, make relationship arrows thicker'"

# Common editing commands
gemini --yolo "/edit 'Add a legend in the bottom right corner'"
gemini --yolo "/edit 'Change background to light gray'"
gemini --yolo "/edit 'Add Salesforce Lightning blue (#0176D3) to headers'"

Editing via Python Script (with resolution control)

uv run scripts/generate_image.py \
  -p "Add legend in bottom right, use SLDS colors" \
  -i ~/nanobanana-output/previous-erd.png \
  -f "erd-with-legend.png" \
  -r 2K

Artistic Styles & Variations

The CLI extension supports style modifiers for creative control:

# Apply artistic styles
gemini --yolo "/generate 'Salesforce ERD' --styles=modern,minimalist"

# Available styles
# photorealistic, watercolor, oil-painting, sketch, pixel-art
# anime, vintage, modern, abstract, minimalist

# Generate multiple variations
gemini --yolo "/generate 'Dashboard mockup' --count=3 --variations=color-palette"

# Reproducible output with seed
gemini --yolo "/generate 'Account ERD' --seed=42"

Workflow Patterns

Pattern A: Visual ERD Generation

Trigger: User asks for visual ERD, rendered diagram, or image-based data model

Workflow:

Run prerequisites check
Query object metadata via sf-metadata (if org connected)
Build Nano Banana prompt with object relationships
Execute Gemini CLI with /generate command (requires --yolo flag)
Open result in macOS Preview app using open command

Example:

# Generate image
gemini --yolo "/generate 'Professional Salesforce ERD diagram showing:
   - Account (blue box, center)
   - Contact (green box, linked to Account with lookup arrow)
   - Opportunity (yellow box, linked to Account with master-detail thick arrow)
   Include legend. Clean white background, Salesforce Lightning style.'"

# Open in macOS Preview (DEFAULT)
open ~/nanobanana-output/[generated-file].png

Pattern B: LWC Mockup

Trigger: User asks for component mockup, wireframe, or UI design

Workflow:

Load appropriate template from templates/lwc/
Customize prompt with user requirements
Execute via Nano Banana
Open in Preview app

Pattern C: Parallel Code Review

Trigger: User asks for code review, second opinion, or "review while I work"

Workflow:

Run Gemini in background with JSON output
Claude continues with current task
Return Gemini's findings when ready

Example:

gemini "Review this Apex trigger for:
   - Bulkification issues
   - Best practices violations
   - Security concerns (CRUD/FLS)
   Code: [trigger code]" -o json

Pattern D: Documentation Research

Trigger: User asks to look up, research, or find documentation

Workflow:

Run Gemini with documentation query
Return findings with sources

Commands Reference

Image Generation

# Generate image from prompt (MUST use --yolo for non-interactive)
gemini --yolo "/generate 'description'"

# Images are saved to ~/nanobanana-output/

Image Display

# Open in macOS Preview app (default)
open /path/to/image.png

# Open most recent generated image
open ~/nanobanana-output/$(ls -t ~/nanobanana-output/*.png | head -1)

# View inline in Claude Code conversation (multimodal vision)
# Use the Read tool → /path/to/image.png

Cross-Skill Integration

| Skill | Integration | Usage | |-------|-------------|-------| | sf-diagram | Enhance Mermaid with visual rendering | "Convert this Mermaid ERD to a visual diagram" | | sf-metadata | Get object/field data for ERDs | Query org before generating ERD | | sf-lwc | Generate component mockups | "Mockup for the AccountList component" | | sf-apex | Review Apex code via Gemini | "Get Gemini's opinion on this trigger" |

Helper Scripts

| Script | Location | Purpose | |--------|----------|---------| | check-prerequisites.sh | scripts/ | Verify all requirements before use | | generate_image.py | scripts/ | Direct API script for 4K resolution and image editing |

Template Usage

ERD Templates (`templates/erd/`)

core-objects.md - Standard CRM objects
custom-objects.md - Custom data model

LWC Templates (`templates/lwc/`)

data-table.md - Lightning datatable mockups
record-form.md - Record form mockups
dashboard-card.md - Dashboard card mockups

Architecture Templates (`templates/architecture/`)

integration-flow.md - Integration architecture diagrams

Review Templates (`templates/review/`)

apex-review.md - Apex code review prompts
lwc-review.md - LWC review prompts

Troubleshooting

Prerequisites Check Failed

Run scripts/check-prerequisites.sh and fix each issue:

No API Key: Set GEMINI_API_KEY in ~/.zshrc (personal key from aistudio.google.com)
No Gemini CLI: Install with npm
No Nano Banana: Install extension via gemini CLI

Image Not Opening

Ensure you're on macOS (Preview app is macOS-only)
Check the file path exists: ls ~/nanobanana-output/
Try opening manually: open ~/nanobanana-output/[filename].png

API Key Errors

Ensure GEMINI_API_KEY is exported in current shell
Verify key is valid at https://aistudio.google.com/apikey
Check billing is enabled on Google Cloud project

Security Notes

⚠️ NEVER commit your GEMINI_API_KEY to version control

Store API key in ~/.zshrc only (not in project files)
The key is personal and tied to your Google account billing

Agent Skills: sf-imagen: Salesforce Visual AI Skill

Skill Files