sf-diagram-nanobananapro: Salesforce Visual AI Skill

Use this skill when the user needs rendered visuals, not text diagrams: ERDs, UI mockups, architecture illustrations, slide-ready images, or image edits using Nano Banana Pro.

Hard Gate: Prerequisites First

Always run the prerequisites check before using the skill:

~/.claude/skills/sf-diagram-nanobananapro/scripts/check-prerequisites.sh

If prerequisites fail, stop and route the user to setup guidance in:

• references/gemini-cli-setup.md

---

When This Skill Owns the Task

Use sf-diagram-nanobananapro when the user wants:

• PNG / SVG-style image output
• rendered ERDs or architecture diagrams
• LWC or Experience Cloud mockups / wireframes
• visual polish beyond Mermaid
• edits to a previously generated image

Delegate elsewhere when the user wants:

• Mermaid or text-only diagrams → sf-diagram-mermaid
• metadata discovery for ERDs → sf-metadata
• LWC implementation after the mockup → sf-lwc
• Apex review / implementation → sf-apex

---

Required Context to Gather First

Ask for or infer:

• image type: ERD, UI mockup, architecture illustration, or image edit
• subject scope and key entities / systems
• target quality: draft vs presentation vs production asset
• preferred style and aspect ratio
• whether the user wants quick mode or an interview-driven prompt build

---

Interview-First Workflow

Unless the user explicitly asks for quick/simple/just generate, ask clarifying questions first.

Minimum question set

Request type	Ask about
ERD / schema	objects, visual style, purpose, extras
UI mockup	component type, object/context, device/layout, style
architecture image	systems, boundaries, protocols, emphasis
image edit	what to keep, what to change, output quality

Question bank: references/interview-questions.md

Quick mode defaults

If the user says “quick”, “simple”, or “just generate”, default to:

• professional style
• 1K draft output
• legend included when helpful
• one image first, then iterate

---

Recommended Workflow

1. Gather inputs

Decide which of these are needed:

• object list / metadata
• purpose: draft vs presentation vs documentation
• desired aesthetic
• aspect ratio / resolution
• whether this is a fresh render or edit of an existing image

2. Build a concrete prompt

Good prompts specify:

• subject and scope
• composition / layout
• color treatment
• labels / legends / relationship lines
• output quality goal

3. Generate a fast draft first

gemini --yolo "/generate 'Professional Salesforce ERD with Account, Contact, Opportunity; clean legend; white background; Salesforce-style colors'"

4. Iterate before final

Use natural-language edits:

gemini --yolo "/edit 'Move Account to center, thicken relationship lines, add legend in bottom right'"

5. Use the Python script for controlled final output

Use the script when you need higher resolution or explicit edit inputs:

uv run scripts/generate_image.py \
  -p "Final production-quality Salesforce ERD with legend and field highlights" \
  -f "crm-erd-final.png" \
  -r 4K

Full iteration guide: references/iteration-workflow.md

---

Default Style Guidance

For ERDs, default to the architect.salesforce.com aesthetic unless the user asks otherwise:

• dark border + light fill cards
• cloud-specific accent colors
• clean labels and relationship lines
• presentation-ready whitespace and hierarchy

Style guide: references/architect-aesthetic-guide.md

---

Common Patterns

Pattern	Default approach
visual ERD	get metadata if available, then render a draft first
LWC mockup	use component template + user context + one draft iteration
architecture illustration	emphasize systems and flows, reduce field-level detail
image refinement	use /edit for small changes before regenerating
final production asset	switch to script-driven 2K/4K generation

Examples: references/examples-index.md

---

Output / Review Guidance

After generating, do one of these:

• open the file in Preview for visual inspection
• attach/read the image in the coding session for multimodal review
• ask the user whether to iterate on layout, labeling, or color before finalizing

Keep the first pass cheap; only spend on high-res output after the composition is right.

---

Cross-Skill Integration

Need	Delegate to	Reason
Mermaid first draft or text diagram	sf-diagram-mermaid	faster structural diagramming
object / field discovery for ERD	sf-metadata	accurate schema grounding
turn mockup into real component	sf-lwc	implementation after design
review Apex / trigger code in parallel	sf-apex	code-quality follow-up

---

Reference Map

Start here

• references/gemini-cli-setup.md
• references/interview-questions.md
• references/iteration-workflow.md

Visual style / examples

• references/architect-aesthetic-guide.md
• references/examples-index.md
• assets/erd/
• assets/lwc/
• assets/architecture/
• assets/review/

---

Score Guide

Score	Meaning
70+	strong image prompt / workflow choice
55–69	usable draft with iteration needed
40–54	partial alignment to request
< 40	poor fit; re-interview and rebuild prompt