30-design-master-global

Design Master — 30-design-master-global

A single master skill that handles 8 design categories for ai-business-skills. One door for both personal work (founder, creator) and brand work (logo, campaign, day-to-day marketing). Reads brand identity from the project context, composes prompts in the right brand voice, and routes the request to the right tool: direct generation via gpt-image-2, paste-ready prompts for 5 popular platforms, or dispatch via Open Design infrastructure.

The skill is written in plain professional English so a non-designer client can read it, understand it, and fill it in alongside the AI. Frontmatter field names stay in English (do not translate) to keep AI parsing intact.

When the user provides a reference image or asks for prompts for image/video/storyboard generation, act as a Prompt Director: classify intent, analyze the image into prompt layers, output an English copy-paste prompt, and state which assets are needed for each image.

The 8 design categories

When to trigger / when NOT to trigger

Triggers (invoke this skill)

• "design image", "create logo", "campaign poster", "campaign banner"
• "social post design", "infographic", "key visual", "hero web mockup"
• "marketing visual", "personal avatar", "monogram", "quote graphic"
• "MidJourney prompt", "DALL-E prompt", "Leonardo prompt", "Imagen prompt"

DO NOT INVOKE (route to a different skill)

Mode detection cascade (Step 0)

Three layers of type detection, in priority order:

Layer 1 — Keyword auto-detect

Layer 2 — Explicit flag

The user passes --type=business-logo (or one of the other 7 types) → overrides Layer 1.

Layer 3 — Ambiguous fallback

If nothing matches and no flag is set → ask one single question:

"Which of the 8 design types is this?

1. personal-brand 2) business-logo 3) business-campaign 4) marketing-day-to-day 5) editorial 6) infographic 7) web-mockup 8) quote-graphic"

Print a transparency line before continuing:

[detect type: business-logo — matched keyword "logo" + brand name "Lattéa"]

Prompt Director mode

Activate this mode when the user provides reference images, asks for a master prompt, wants face/color/logo replacement, or gives feedback on a failed result. Load references/prompt-director.md before composing the prompt.

Rules:

• Do not start with a long questionnaire. If the request is usable, generate the prompt first and list missing pieces at the end.
• A reference image is a style/composition source, not automatically a face, logo, or product source unless the user says so.
• Ask for replacement assets by type: face reference, logo file, brand palette hex, product image, background/location, exact text.
• Multiple images = multiple flows. Name them Flow A/B/C, state each image role, provide a prompt, a negative prompt, and the missing-asset list for each flow.
• Final prompts for image/video models default to English; user-facing explanation follows the user's language.

Reference-image output skeleton:

## Image flows

| Flow | Input image | Role | Output |
|------|-------------|------|--------|
| A | [file/path] | style/composition reference | master prompt |

## Copy-paste prompt — Flow A
[English prompt]

## Negative prompt / Avoid
[English avoid list]

## Upload next to personalize Flow A
- Face reference:
- Logo:
- Brand colors:
- Exact text:
- Product/background:

Runtime workflow — 5 steps

Step 1 — DETECT design type

Run the 3-layer cascade above. Print [detect type: ...] before moving to Step 2.

Step 2 — LOAD references

Load lazily, as needed:

• CONVENTIONS.md — load once per session
• references/prompt-director.md — load for reference images, master prompts, face/color/logo replacement, multiple images, or prompt feedback
• references/<type>.md — load the file matching the type just detected (e.g. references/business-logo.md)
• references/brand-identity-source.md — load whenever the type starts with business-*
• references/fallback-prompt-format.md — load when the tier is Free (see Step 5)
• templates/<format>.md — load the template matching the requested format (e.g. templates/poster.md, templates/social-square.md)

Step 3 — READ brand identity

Business modes (business-logo / business-campaign / marketing-day-to-day): REQUIRED — must find brand identity. Auto-search in priority order:

1. assets/brand/logo.{svg,png,ai} — primary logo file
2. brand-guideline.md / brand-identity.md at the project root
3. 02-campaign-brief-global/output.md section "Brand mandatories" — campaign brief carries brand constraints
4. prd.md frontmatter target_user + section "Visual cues" — style direction
5. prd.md Section 6 — features that inform the visual concept

If nothing is found for a business mode → BLOCK + ask the user to upload a logo + specify a palette (3 hex codes) + name a primary font family.

Personal modes (personal-brand / personal quote-graphic): brand identity is optional. Enough to have style adjectives (3–5 words) + a color preference (3 mood colors).

Editorial / Infographic / Web-mockup: context-dependent — treat as business mode if the asset is for a brand, treat as personal if it's for an individual or a general-purpose piece.

Step 4 — ROUTE by type

Step 5 — CHECK API tier + GENERATE

Detect the tier in Bash:

if [[ -n "$OPENAI_API_KEY" ]]; then
  TIER="pro"   # direct gpt-image-2
elif [[ -n "$OD_BIN" && -x "$OD_BIN" ]]; then
  TIER="enterprise"   # Open Design dispatcher
else
  TIER="free"   # prompt-only fallback
fi

Print line: [tier: <X>] [model: <Y>].

Type 7 web mockup — after generating the hero image (if Pro/Enterprise tier), print a recommendation block:

[NEXT STEP] Hero image generated. For a full multi-section interactive mockup,
also run one of:
  - web-prototype (general landing/marketing page)
  - saas-landing (SaaS product page)
  - mobile-app (mobile app screens)
  - dashboard (admin/analytics dashboard)
  - frontend-design (Figma-style design tokens)

Type 8 quote graphic — output 2 files:

• docs/design/<slug>-bg.png — background image generated via gpt-image-2 (NO text)
• docs/design/<slug>.html — HTML overlay rendering the quote text on top of the background (HTML keeps typography crisp; avoids AI garbling letters)

Frontmatter schema for the output file

Every docs/design/<slug>.md metadata file MUST carry this frontmatter:

---
title: <Image asset name>
type_artifact: image-generated | image-prompt-fallback | infographic-html | web-hero-image
mode: personal-brand | business-logo | business-campaign | marketing-day-to-day | editorial | infographic | web-mockup | quote-graphic
subject: <1-line description>
source_brief: <path to source brief if any>
brand_identity:
  logo_path: <path or "none">
  palette: [<hex>, <hex>, ...]
  typography: <font family or "none">
  style_adjectives: [<adj>, <adj>, ...]
format: poster | social-square | social-vertical | banner-hero | magazine | infographic | logo-variant | quote
aspect_ratio: 1:1 | 9:16 | 16:9 | 3:4 | 4:3 | custom
gen_mode: api-direct | api-dispatcher | fallback-prompt | hybrid
model: gpt-image-2 | dall-e-3 | midjourney-v6 | flux | imagen-3 | manual
output_files:
  - <path>
created: 2026-05-20
last_updated: 2026-05-20
---

Field names stay in ENGLISH — do not translate (would break AI parsing). The body of the file can be in any language the client reads.

Error handling — 8 real-world situations

Anti-patterns — DO NOT

• Generate a logo as the final deliverable → always produce a multi-variant set (3–5 directions) + human-review disclaimer
• Burn >50 characters of text into an image — AI gen mangles long text, breaks spelling, distorts accented characters
• Pass a raster web mockup off as interactive — must be hybrid + recommend an HTML skill
• Skip the brand identity check on business modes — the output will look disconnected from the brand
• Output a prompt that's just a keyword list (e.g. "nice cafe logo") — must follow the 5-section structured prompt framework: subject + composition + lighting + palette + style + negatives
• Hardcode hex colors when the project already has a brand identity file — read from source
• Translate frontmatter field names to a non-English language — breaks AI parsing
• Generate >1 image per turn, unless it's a logo multi-variant set
• Save only a prompt without generating the image while on Pro/Enterprise tier — wastes the API budget
• Generate an image without saving a .md metadata sidecar — context is lost next time

Self-test before generating

Before every gpt-image-2 call or fallback prompt export, ask yourself:

"Does this output align with the brand voice and identity? Would a client recognize the brand when they see it? If this is personal brand work — does it reflect the style adjectives the user actually wrote down?"

If you can't confidently answer 2 out of 3 → go back to Step 3, re-read the brand identity, or probe the user with 1–2 more questions.

Examples reference

• examples/personal-founder-avatar.md — Founder avatar for OPA (style: confident, warm, minimal)
• examples/lattea-fall-campaign-poster.md — Key visual for the Lattéa fall campaign (palette: brown + burnt orange)
• examples/ai-income-mastery-quote-graphic.md — IG quote graphic for the "AI Income Mastery" course (background + HTML overlay)
• examples/opa-roadmap-infographic.md — Product roadmap infographic for OPA (icons-heavy, text minimal)

Cheat sheet — which file to load and when

Plain English policy

This skill is written in plain professional English so a non-designer business owner can read it, follow along, and fill it in alongside the AI. Technical terms (gpt-image-2, DALL-E 3, MidJourney) are kept verbatim and briefly explained the first time they appear. Examples use realistic global-feeling brands: Lattéa (a fictional global cafe chain), Lumière Studio (creative agency), AI Income Mastery (online course), OPA (product). No throwaway placeholder brand names.

Self-test for plain-English fit: "Would a small business owner who doesn't speak design jargon understand this section?" If not — rewrite it.