Use when extracting a production-ready brand identity and design system from curated reference images, mood boards, or brand descriptions. Triggers include requests to "extract brand identity", "create brand design system", "define brand DNA", "generate design tokens from images", "build a brand from moodboard", "create AI image prompts from brand", "design AI prompt methodology for brand", or "systematize brand identity for front-end handoff". Also triggers when the user provides reference images and asks for a structured brand output or AI image generation prompts.
Use when extracting a production-ready brand identity and design system from curated reference images, mood boards, or brand descriptions. Triggers include requests to "extract brand identity", "create brand design system", "define brand DNA", "generate design tokens from images", "build a brand from moodboard", "create AI image prompts from brand", "design AI prompt methodology for brand", or "systematize brand identity for front-end handoff". Also triggers when the user provides reference images and asks for a structured brand output or AI image generation prompts.
Brand Identity Extractor
Extract a complete, production-ready Brand Design System + Front-end Handoff Kit + AI Image Generation Asset Kit + Prompt Design Methodology from curated reference images, text descriptions, or mood boards.
Three stages: Define → Handoff → Generate
Core Contract
Extract only what the inputs support — do not invent
Produce all 3 stages — skipping any section is a failure
Include both affirmative rules (DO) and negative constraints (DON'T)
Output design tokens in copy-paste-ready JSON
Translate brand into concrete AI image generation prompts, not just adjectives
Produce a prompt design methodology that teaches how to design brand-consistent AI image prompts — the "how to think" layer above the templates
Phase 0: Brand Discovery & Conflict Resolution
Before extraction begins, establish the brand's foundational context. Skip this phase only if all 6 Brand Questions are already answered in provided inputs, or if the user is extending an existing brand with a pre-validated .agents-stack/reference/design.md.
0.1 Source Discovery (Do This Before Asking Any Question)
Scan existing project documents for brand inputs before asking the user anything:
Check in order:
.agents-stack/reference/design.md → product intent, behavior, brand direction
Mood boards, design inspiration, "what we are/aren't"
Rule: If a question is answered in docs, use it — label the source as 【From: filename.md】. Only ask for what genuinely cannot be found. Ask up to 3 clarifying questions per round; if the user still cannot answer, proceed with labeled 【Assumption: ...】.
0.2 The 6 Brand Questions
Q1 — Purpose & Worldview(Brand DNA foundation)
"Why does this brand exist beyond making money, and what change in the world are you obsessed with creating?"
Collect: 1–2 paragraphs. Must contain a clear enemy/problem and the better future it fights for. Drives overall visual tone — activist/bold vs. gentle/nurturing.
Q2 — Audience Empathy(Persona)
"Describe your ideal customer as a real person — name, age, daily frustrations, dreams."
Collect: 150–300 word character sketch. Drives component roundness, color saturation, icon style, density.
Resolve all conflicts silently. Never output tokens before completing all 5 checks.
Check 1 — Vibe Check (Q4 vs Q6): Do Personality adjectives match Visual References' aesthetic? Conflict rule: Visual References govern physics (layout, radius, shadow, density); Personality governs voice (microcopy, motion energy, empty states).
Check 2 — Emotional Territory Check (Q5 vs Q4/Q6): Do Q5 "forbidden words" describe any current direction? If yes → adjust palette or radius before output.
Check 3 — Component Rationalization (Q3): Does at least one planned component directly prove the "Only We" claim? If none → add a Hero Component and flag it.
Check 4 — Persona Compatibility Check (Q2): Does token density and radius suit the persona's world? Design-literate → moderate density, editorial radius. Consumer → more air, softer radius.
Check 5 — WCAG AA Pre-check: Text on surface: minimum 4.5:1 body, 3:1 large text. Fail → adjust token before output. Never output a failing contrast.
AI generation: Midjourney, DALL-E, Stable Diffusion
If no images are given, rely on textual input. Ask clarifying questions only when too vague.
Phase 0 handles all brand discovery and conflict resolution. If inputs are already complete with answers to all 6 Brand Questions, Phase 0 auto-detects this and skips to extraction.
Step 2: Run the Brand Extract Meta-Prompt
Copy the meta-prompt from brand-extract-meta-prompt.md and paste it into a reasoning model (O1, DeepSeek, Claude) with the user's inputs. The meta-prompt produces all 3 stages in one pass.
Step 3: Validate Completeness
After receiving the output, verify all 10 output sections are present:
If any section is missing or sparse, prompt the model again with specific gaps.
Step 4: Cross-check Token Validity
For extracted hex values and numeric tokens, verify against design-standards:
Body text ≥ 16px
Button/input height ≥ 44px (tap target)
Contrast ratios (4.5:1 normal text, 3:1 large text)
Spacing scale (systematic, not arbitrary)
AI prompt negative keywords match Stage 1 forbidden lists
Step 5: Write to .agents-stack/reference/design.md
design.md is the single canonical design reference for the project (per AGENTS.md). All brand identity content goes here. External artifacts are referenced from it — never duplicated.
Write the Visual Universe as a YAML block under ## Visual System:
Stage 1 B.1–B.5 (Color Philosophy, Form Language, Material Library, Composition DNA, Object Library) must be written as a single structured YAML block inside ## Visual System, following the v2.0 brand identity schema. The YAML block must include: color_policy, design_tokens (spacing, radius, shadow, blur, motion, typography), form_language, material_language, scene_density_rules, object_library, ui_translation, negative_prompt_policy, input_variables, application_presets, prompt_seed, and rule_severity. See .agents-stack/reference/design.md for the canonical schema structure.
Prompt Templates → .agents-stack/reference/design/ai-prompt-templates.md (5 templates from Stage 3.4)
Tool-Specific Parameters → .agents-stack/reference/design/ai-prompt-params.md (from Stage 3.3)
Mark any inferential gaps. If a section's content is entirely covered by an external file, write a one-paragraph summary in design.md with the reference link.
Cross-Skill Integration
These skills work well with BIE in a loose pipeline — no hard dependency, just recommended sequencing:
Feed BIE's AI prompt output (Core Style, Negative Bank, templates) into prompt-augmentation with text-to-image mode for domain-specific term substitution (optics, lighting, composition, materials).
BIE's output in .agents-stack/reference/design.md (YAML block under ## Visual System for machine-readable tokens, prose sections for human context) is consumed by design-context-scout as a design system source for UI sprint planning.
Bundled Resources
Meta-Prompt — copy-pasteable 3-stage prompt for reasoning models
