name: sf-diagram-nanobananapro
description: >
AI-powered visual content generation for Salesforce development.
Generates ERD diagrams, LWC mockups, architecture visuals using Nano Banana Pro.
Also provides Gemini as a parallel sub-agent for code review and research.
license: MIT
metadata:
version: "1.5.0"
author: "Jag Valaiyapathy"
scoring: "80 points across 5 categories"
hooks:
PreToolUse:
- matcher: Bash
hooks:
- type: command
command: "python3 ${SHARED_HOOKS}/scripts/guardrails.py"
timeout: 5000
PostToolUse:
- matcher: Bash
hooks:
- type: command
command: "python3 ${SHARED_HOOKS}/suggest-related-skills.py sf-diagram-nanobananapro"
timeout: 5000
SubagentStop:
- type: command
command: "python3 ${SHARED_HOOKS}/scripts/chain-validator.py sf-diagram-nanobananapro"
timeout: 5000

sf-diagram-nanobananapro: 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-diagram-nanobananapro/scripts/check-prerequisites.sh

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

Requirements

Optional (for 4K/editing via Python script)

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)
- Default style: architect.salesforce.com aesthetic (see below)

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

🎨 Default Visual Style: Architect.salesforce.com

All ERD diagrams now default to the official Salesforce architect.salesforce.com aesthetic.

The Signature Look: Dark Border + Light Fill

Cloud-Specific Colors (Auto-Detected)

Standard Elements

See resources/architect-aesthetic-guide.md for full specification.

🔄 Draft → Final Workflow (Cost-Effective Iteration)

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

Recommended Workflow

1. Draft at 1K - Generate initial image quickly
2. Iterate - Refine prompt or use /edit to adjust
3. 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"

🎤 Interview-First Workflow (Recommended)

Before generating any image, Claude MUST ask clarifying questions using AskUserQuestion.

This ensures high-quality output by gathering requirements upfront rather than guessing.

Automatic Interview Triggers

Interview Flow

User: "Generate an ERD for my org"
        │
        ▼
┌─────────────────────────────────────────────────────────┐
│  Claude invokes AskUserQuestion with 4 questions:       │
│                                                         │
│  Objects: [Core CRM / Sales Cloud / Service Cloud / Custom]  │
│  Style:   [Professional / Whiteboard / Technical / Minimalist]│
│  Purpose: [Documentation 4K / Quick draft 1K / Presentation]  │
│  Extras:  [Legend / Field names / Color-code] (multi-select)  │
└─────────────────────────────────────────────────────────┘
        │
        ▼
Claude builds optimized prompt from answers
        │
        ▼
Generate image with gathered requirements

Example Questions for ERD

Question 1: "Which objects should be included in the ERD?"
- Core CRM (Account, Contact, Opportunity, Case) ← Recommended
- Sales Cloud (Lead, Campaign, Quote, Order)
- Service Cloud (Case, Knowledge, Entitlement, Asset)
- Custom objects (I'll specify)

Question 2: "What visual style do you prefer?"
- Professional (clean lines, SLDS colors) ← Recommended
- Whiteboard (hand-drawn, casual)
- Technical (with field names)
- Minimalist (simple boxes)

Question 3: "What's the primary purpose?"
- Documentation (4K quality)
- Quick draft (1K, fast iteration) ← Recommended for first attempt
- Presentation slides
- Architecture review

Question 4: "Any special requirements?" (multi-select)
- Include legend
- Show field names
- Color-code by object type
- None

Skip Interview (Quick Mode)

To skip questions and use defaults, include these keywords:
- "quick" → "Quick ERD of Account-Contact"
- "simple" → "Simple LWC mockup"
- "just generate" → "Just generate an architecture diagram"

Quick mode uses: Professional style, 1K resolution, legend included.

Full Question Reference

See docs/interview-questions.md for:
- Complete question sets for all image types
- Answer-to-prompt mapping tables
- Default values for skip mode

Workflow Patterns

Pattern A: Visual ERD Generation

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

Workflow:
0. Run Interview (unless "quick" mode) - Ask about objects, style, purpose, extras
1. Run prerequisites check
2. Query object metadata via sf-metadata (if org connected)
3. Build Nano Banana prompt using interview answers + object relationships
4. Execute Gemini CLI with /generate command (requires --yolo flag)
5. 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:
0. Run Interview (unless "quick" mode) - Ask about component type, object, context, style
1. Load appropriate template from templates/lwc/
2. Customize prompt using interview answers + template
3. Execute via Nano Banana
4. Open in Preview app

Pattern C: Parallel Code Review

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

Workflow:
1. Run Gemini in background with JSON output
2. Claude continues with current task
3. 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:
1. Run Gemini with documentation query
2. 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

Helper Scripts

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

License

MIT License. Copyright (c) 2024-2025 Jag Valaiyapathy