Brand Workshop

Brand Workshop

A collaborative branding skill that assembles a virtual creative team, runs a structured brainstorming session, and delivers a brand strategy brief, tagline, and logo.

STOP — When NOT to use this skill

This skill generates brand identity from a business overview. It is NOT a remix tool for a brand that already exists. Hand off to team-composer — do not run the workshop — if any of these are true:

• The user already has a logo in use and is asking to refresh / update / audit it
• The user already has a style guide and is asking to modernize or refresh it
• The user says "our current brand", "our existing voice", "our logo", "our palette"
• The request is "update our brand voice" / "refresh our style guide" / "our brand has evolved"
• The deliverable is a style-guide update, voice audit, or brand refresh — not a net-new identity

How to hand off:

Invoke team-composer:team-composer as the next skill. Use the literal skill identifier — do NOT pick a plausibly-adjacent skill by name match alone.

Anti-patterns — do NOT invoke any of these in place of team-composer:

• brand-voice:guideline-generation — this is for extracting voice from existing sales calls / uploaded brand documents. It's a voice-discovery workflow, not a refresh workshop. Using it here will pull the user into a discovery loop they didn't ask for.
• brand-voice:discover-brand — this searches Notion / Slack / Drive for scattered brand materials. The user already knows where their brand lives — they want to evolve it, not find it.
• brand-voice:brand-voice-enforcement — this applies an already-written voice guide to new content. It does not produce a refresh.
• brand-workshop:brand-workshop (this skill) — do not recursively re-enter after the STOP check.

Once inside team-composer:team-composer, use @brand_strategist + @senior_copywriter for voice work, or @naming_specialist for naming. Do not proceed with the Discovery → Concept → Creation workshop described below.

Overview

This skill simulates a brand workshop with specialized roles working together. The process moves through three phases: Discovery → Concept → Creation. The output is always:

1. Brand Strategy Brief (brand-brief.md) — the team's rationale, concepts explored, and final direction
2. Tagline — embedded in the brief and called out prominently
3. Logo (.svg rendered to .png) — code-generated, defaulting to minimalist style
4. Favicon Pack — favicon.svg, raster sizes (16/32/180/512), site.webmanifest, and a copy-paste HTML <link> install snippet
5. Social Banner Set — Open Graph (1200×630), X header (1500×500), LinkedIn banner (1584×396), Instagram square (1080×1080), profile avatar (400×400)
6. Descriptions Pack (descriptions.md) — tagline + short/medium/long bios + elevator pitch + press boilerplate, all matched to brand voice
7. Starter Design System (DESIGN.md) — tokens only: color, typography, spacing, radius, and voice principles (no component specs)

Scope boundary: pitch decks (both the deck template and filled slide content) and the Business Model Canvas are intentionally out of scope. The pitch-deck skill reads DESIGN.md directly and generates a brand-skinned deck on its own — brand-workshop does not pre-emit a deck template. See Skill Boundaries below.

Skill Boundaries

This skill overlaps partially with team-composer. Both assemble a virtual team and run structured discussion. The difference is what they produce.

Use team-composer instead of this skill when:

• The user wants general project brainstorming, planning, or review — even if branding is part of it
• The user wants to name a product/feature (use @naming_specialist), discuss positioning (use @brand_strategist), or review brand voice (use @humorist + @senior_copywriter)
• The deliverable is NOT a logo + tagline + brief package
• The user is making multi-dimensional product decisions where branding is one input among many

Stay in this skill when:

• The primary deliverable is a brand identity package: logo (SVG) + tagline + brand strategy brief
• The user says: "help me brand my [startup/product/app]", "I need a logo", "create a brand identity", "design a brand concept", "give me a tagline and logo"
• The user provides a business overview and asks for visual identity work

Boundary examples:

Stop-rule — existing brands:

If the user's request is to audit, refresh, document, or extend an existing brand (logo already in use, style guide already drafted, "make our current brand more X", "update our brand voice"), stop and invoke team-composer instead. This skill generates brand identity from a business overview — it is not a remix/audit tool for a live brand.

Future refactor note (Option D): Today this skill has its own internal team (see Phase 2). Long-term, this skill could invoke team-composer as the discussion engine and focus only on the Creation phase (logo SVG, brief assembly). That refactor is not scheduled — revisit only if the role catalog drift between the two skills causes real maintenance pain. When revisited, unify role tag format (@hyphen-case here vs. @snake_case in team-composer).

Companion plugins (the startup pipeline). This skill is step 1 of 5: brand-workshop → validation-canvas → riskiest-assumption-test → pitch-deck → startup-grill. Each lives in its own plugin directory. They are intentionally not folded together: each requires founder inputs (traction, revenue model, ask size, test results) that are out of scope for a brand workshop. After this skill ships, suggest validation-canvas as the next step — see Phase 7 (Closing). An umbrella startup-launch-kit plugin may later orchestrate all five.

Phase 1: Discovery

Step 0.0 — Path resolution + manifest awareness (v2.2.0+)

Resolve the brand root once at invocation, in this precedence order (canonical chain):

1. Explicit output_dir arg (passed by startup-launch-kit) → use as-is.
2. STARTUP_KIT_DOCS_ROOT env var set → ${STARTUP_KIT_DOCS_ROOT}/brand/.
3. Smart default — docs/startup-kit/ exists → docs/startup-kit/brand/. Surface the smart-default notice: "Writing to docs/startup-kit/brand/ (smart default). Set STARTUP_KIT_DOCS_ROOT=./docs to write standalone instead."
4. Solo fallback → docs/brand/.

Note (v2.2.0 path migration): prior versions wrote artifacts to the working-directory root and downstream skills expected brand-kit/ as the prefix — these were inconsistent. v2.2.0 standardizes on <brand-root>/ (default docs/brand/). Downstream skills include legacy brand-kit/ fallback reads for backward compat.

Manifest awareness. Look for kit-manifest.json at <resolved-kit-root>/kit-manifest.json first; fall back to the working-directory root for backward compat. Use it as a hint, never as a bypass:

• If the manifest lists brand-workshop as completed with a recent mtime, surface that fact: "Manifest says you ran brand-workshop on [date]. Update mode (revise specific assets), fresh run (rebuild everything), or skip to the next step?"
• If the manifest's intake_answers cache is populated, the founder has run the orchestrator's intake — use those answers as defaults for the discovery questions below (still surface them; do not silently skip).
• Manifest read failures (corrupt JSON, missing fields) are non-fatal — log the issue inline and proceed as if no manifest exists.

After this skill ships its artifacts (Phase 6 — output), if a manifest exists, append/update this skill's entry. Use atomic write (write .tmp, then rename). If the manifest doesn't exist, do NOT create it — that's the startup-launch-kit orchestrator's job. See startup-launch-kit/references/manifest-schema.md for the schema.

Gather Inputs

Collect the following from the user. If some are missing, ask concisely — but don't block on everything. Make reasonable assumptions and state them.

Style Default

If the user does not specify a logo style, default to minimalist with negative space. This style emphasizes clean geometry, hidden meaning through negative space, and timeless simplicity. It works well across digital and print.

If the user asks for style suggestions, read references/logo-styles.md and present the options with brief descriptions and famous examples.

Phase 2: Concept — The Workshop

Assemble the virtual team and run a structured brainstorming. Each role contributes a distinct perspective. The workshop is NOT a free-for-all — it follows a deliberate sequence.

The Team

Read references/team-roles.md for detailed role definitions. Summary:

Not every role needs to speak on every project. Scale participation to the project:

• Minimum (simple logo request): strategist, copywriter, visual designer
• Full team (app/product branding): all roles
• Add roles if context demands it (e.g., @accessibility-specialist for inclusive branding)

Workshop Flow

Run these steps sequentially. Write each role's contribution in first person with their role tag (e.g., **@senior-brand-strategist:**). This becomes the brainstorming transcript in the brand brief.

Step 1: Strategic Foundation (Strategist)

• Define brand positioning: what makes this distinct?
• Identify 2-3 brand archetypes that fit
• Propose core brand attributes (3-5 adjectives)

Step 2: Verbal Exploration (Copywriter)

• Generate 5-7 tagline candidates based on positioning
• Note tone of voice direction
• Flag any naming/linguistic concerns
• Draft the descriptions pack variants (short / medium / long bio, elevator pitch, boilerplate) — these feed directly into Phase 3 and must match the tagline's voice. Each variant should stand alone, not read as a truncation of the next.

Step 3: Visual Direction (Visual Designer + UX/UI Designer)

• Propose 2-3 logo concepts (described verbally — shape, symbolism, approach)
• Suggest color palette direction (with hex values)
• Recommend typography family
• UX/UI designer weighs in on digital viability (favicon test, dark/light modes)

Step 4: Cultural & Market Check (Cultural Advisor + Growth Strategist)

• Cultural advisor flags any issues with symbols, colors, or words across cultures
• Growth strategist evaluates memorability and market positioning

Step 5: Convergence (Strategist leads)

• Synthesize: pick the winning tagline and logo direction
• State the rationale clearly
• The team votes (brief inline votes from each role)
• Declare the final concept

Phase 3: Creation

Brand Strategy Brief

Output a .md file with this structure:

# Brand Strategy Brief: [Business Name]

## Executive Summary
One paragraph: what was created and why.

## Inputs
- Business overview, vision, mission (as provided)

## Positioning
2–4 sentences naming the target segment, the category the brand plays in, and
the one axis of differentiation. This is the section downstream plugins
(`pitch-deck`) parse for the problem/solution narrative — keep it short and
literal, not aspirational.

## Voice & Tone
3–5 bullets from the copywriter describing how the brand speaks (e.g.,
"confident but not arrogant", "plain language over jargon", "contractions OK").
Must match the bullets written into `DESIGN.md → Voice` (the custom Voice
section that brand-workshop appends after Do's and Don'ts) — these are the
same ruleset, surfaced in two places.

## Workshop Transcript
The full brainstorming from Phase 2, role by role.

## Final Concept

### Tagline
> "[The chosen tagline]"

Rationale: Why this tagline wins.

### Logo Concept
Description of the chosen logo direction, symbolism, and style.

### Color Palette
| Color | Hex | Usage |
|-------|-----|-------|
| Primary | #XXXXXX | ... |
| Secondary | #XXXXXX | ... |
| Accent | #XXXXXX | ... |

### Typography
Recommended font families and usage.

## Rejected Alternatives
Brief note on other tagline/logo concepts explored and why they were passed over.

Section-contract note: ## Positioning and ## Voice & Tone are named anchor headings — pitch-deck greps by them. If you rename, update pitch-deck/SKILL.md Phase 1 Step 1 in lockstep.

Logo Creation

Generate the logo as SVG using Python. Follow these principles:

Design Principles (Minimalist / Negative Space default):

• Clean geometric shapes — circles, rectangles, triangles, arcs
• Negative space to embed hidden meaning (e.g., the arrow in FedEx, the bear in Toblerone)
• Maximum 2-3 colors from the chosen palette
• The logo must work at 32x32 (favicon) and 512x512
• No raster effects — pure vector geometry
• Text in the logo (if any) should use basic sans-serif or be drawn as paths
• Aim for a logomark (icon) + optional logotype (wordmark) lockup

Technical approach:

1. Generate SVG using Python string construction or svgwrite library
2. Keep the SVG clean — no unnecessary groups or transforms
3. Render to PNG at multiple sizes (64px, 256px, 512px) using cairosvg or librsvg
4. If rendering tools aren't available, output the SVG and note that PNG conversion can be done externally

Code quality:

• Write the SVG generation code clearly with comments explaining each shape's purpose
• Test the SVG is valid XML before saving
• Include viewBox for proper scaling

If the user requests AI-generated imagery instead: Use the Hugging Face dynamic_space tool to find and invoke an image generation space. This is opt-in only — never default to it.

Favicon Pack

Generate a favicon set from the logomark. The mark must remain legible at 16×16 — if fine detail disappears at that size, author a simplified favicon-optimized variant (same concept, fewer strokes) rather than scaling the full mark blindly.

Outputs (under favicons/):

install snippet template:

<link rel="icon" type="image/svg+xml" href="/favicon.svg">
<link rel="icon" type="image/png" sizes="32x32" href="/favicon-32.png">
<link rel="icon" type="image/png" sizes="16x16" href="/favicon-16.png">
<link rel="apple-touch-icon" sizes="180x180" href="/favicon-180.png">
<link rel="manifest" href="/site.webmanifest">
<meta name="theme-color" content="#XXXXXX">

Quality rule: Programmatically (or mentally) test the 16×16. If the mark is unrecognizable at that size, simplify before generating — do not ship an illegible favicon.

Mascot Lane (Generative Raster) — optional

The mascot is the one brand asset that is artwork, not spec. Route accordingly:

The favicon / logo head-mark never derives from the mascot raster — it stays in the authored SVG lane, always.

Style resolution (from Discovery). Default style: flat sticker illustration — bold clean outlines, flat colors, white sticker border, plain ground-tint background. A default, not a rule: resolve from Discovery signals (style preference, audience, positioning) — a retro-positioned brand may choose pixel-native instead (a separate pixel-art generation with its own locked recipe).

Generation flow:

1. Anchor-steered prompt. Brand hexes enter the prompt as anchors (steering), never a hard spec — name 3–5 palette roles with hexes ("<ground> fur, <brand-dark> markings, <accent> prop"). Acceptance is the palette-anchor gate in the checklist, not exact color reproduction.

2. Master generation. One character, seed-locked, ≥1024px. The canon pose IS the character.

3. Write mascot.md immediately — the recipe is the master; the PNG is just one render. Schema (example: references/mascot.md.example):

   • generator: id and version/space id (seed reproducibility is generator-version-bound)
   • seed: · resolution:
   • LOCK: one paragraph — the character description that never changes
   • POSE: pattern — a new pose appends one POSE: <one line> to the locked prompt; change nothing else
   • DERIVE: lanes — cutout → transparent · vtracer → SVG (scale/print utility) · quantize → pixel (quick retro moments)
   • Known limits — cross-pose generative drift; the canon master is authoritative

4. Poses (optional). Append POSE: lines per the recipe. Expect drift; regenerate rather than accept an off-model pose.

5. Cutout — python3 <skill-dir>/scripts/verify.py cutout <brand-root>/mascot/master.png <brand-root>/mascot/master-transparent.png: edge flood-fill from corner-median background (default --tol 28, tuned against the white sticker border), drops pale low-saturation components (soft shadows), keeps saturated satellites (hearts, sparkles), feathers alpha.

   verify.py command convention: verify.py lives in this skill's own directory and is not copied into the output; the assets live under the resolved <brand-root>/mascot/ in the user's project. So the script path and the asset paths come from two different roots — resolve <skill-dir> as the directory this SKILL.md was loaded from (in this source repo skills/brand-workshop; when installed as a plugin, the installed skill path — do not assume the current project has a skills/… tree), and <brand-root> as the resolved brand root in the user's project (e.g. docs/brand). Both must be real paths in the same command; the bare scripts/verify.py … / mascot/… shorthand is not runnable.

6. Character sheets. Instantiate templates/character-sheet-web.html.template and templates/character-sheet-print-landscape.html.template (print default = landscape; portrait is an optional variant). Token tables live in each template's header comment. Fill by string replace only — never str.format (format-style brace doubling shipped a half-broken sheet once; the integrity gate exists because of it). HTML-escape every scalar copy token before replacing — a brand name or tagline with &, <, >, or " dropped raw into body text or an attribute breaks the sheet (and trips the integrity gate); escape & < > " for text/attribute slots. Row-fragment tokens the copywriter builds as markup (IDENTITY_ROWS, POSE_TILES, CHIPS, STATS_ROWS, …) and base64 image tokens are already markup/data — insert those raw. The copywriter role emits the full token map — names, backstory, stats labels, voice lists, mock copy — as a single map; templates carry structure only: no content logic, no fallback copy.

Derivatives are utilities, not heroes. The traced SVG and quantized pixel versions are scale/print and retro conveniences. A hero pixel mascot is a separate native pixel-art generation with accepted drift and its own locked recipe.

Deliverables are the HTML sheets + PNGs (see Output Files). Pagegate PDFs are verification-only artifacts — delete after the gate; they never ship.

No generator reachable? Ship a first-class prompt brief (model-agnostic prompt + the mascot.md recipe skeleton) — mirror pixel-art's Path B. Never downgrade to a code-drawn mascot, and don't call the brief a "fallback" in user-facing output.

Quality amplifiers (optional, referenced by literal ID). Greenfield sheet generation → impeccable (brand register; tested at 3.9.1). Revising an existing sheet → taste-skill:redesign-skill (as of 2026-07-07; recorded at handoff time as redesign-existing-projects — use the ID that resolves in your runtime). When absent, degrade gracefully to this skill's checklist. Never "pick something from the taste-skill plugin" untargeted.

Social Banner Set

Generate branded social assets from the logo, palette, and typography. Use generous whitespace — do not crop the logo to the edge. Banners that look auto-stamped hurt the brand more than helping it.

Outputs (under social/):

Design rules:

• Minimum 8% padding on the shorter axis — logo never touches the edge
• Tagline appears only on hero-style banners (og-image, instagram-square); profile headers stay clean
• Background: brand primary or secondary; fall back to a neutral if contrast fails WCAG AA for any overlaid text
• Generate as SVG, rasterize to PNG via cairosvg (same toolchain as logo)
• If the palette has strong light/dark candidates, ship both variants of x-header and linkedin-banner; minimum is one variant per file

Quality rule: Preview each banner at 25% scale. If the composition looks cramped at full size, it will look worse in a timeline.

Descriptions Pack

The copywriter's work from Phase 2 Step 2 is assembled into a single file. All variants must match the voice established in the brief and stand alone — not read as truncations of each other.

Output: descriptions.md

Quality rule: Paste each variant into the appropriate platform's character counter before shipping. If any variant is over its hard limit, rewrite — do not truncate.

DESIGN.md (Starter Tokens)

Output DESIGN.md as its own standalone file at the root of the output folder, named exactly DESIGN.md (uppercase). When the founder adopts the brand kit into a real repo, this file moves to the repo root — that's the Google Labs DESIGN.md spec convention. Do NOT fold tokens into brand-brief.md as a section — downstream plugins (validation-canvas, riskiest-assumption-test, pitch-deck) parse DESIGN.md directly and will not find tokens buried in another file.

Format: Google Labs DESIGN.md, spec version alpha (current as of 2026-05; the spec is under active development — re-check when the spec ships a non-alpha tag). The file has two parts:

1. YAML front matter — machine-readable design tokens.
2. Markdown body — human-readable rationale, organized into canonical sections.

Scope is deliberately narrow: tokens, not components. Button styles, form fields, grids, motion, and elevation depend on the engineering stack this skill does not choose — the implementing team adds those when they wire up the design system in their framework.

Cross-plugin contract. Downstream plugins (validation-canvas, riskiest-assumption-test, pitch-deck) bind their accent CSS variable (--canvas-accent, --rat-accent, --deck-accent) to colors.primary from the YAML front matter. Use the spec-recommended token names — primary, secondary, tertiary, neutral — do not invent alternatives or rename the keys. The previous Primary | Secondary | Accent convention is retired in favor of spec-aligned naming; downstream plugins read the YAML directly rather than greppping prose labels.

Section order (per spec). The DESIGN.md spec defines a canonical section order. Brand-workshop emits a subset for "starter" scope. Sections present must appear in this order:

1. Overview (required) — brand personality, audience, intended emotional response
2. Colors (required) — palette rationale + per-color usage prose
3. Typography (required) — type strategy + per-level role
4. Layout (required, even if minimal) — grid model + spacing rhythm
5. Shapes (required) — corner-radius philosophy
6. Do's and Don'ts (required) — visual / design guardrails
7. Voice (custom section appended after Do's and Don'ts; preserved by spec consumers as unknown but valid)

Sections explicitly omitted at "starter" scope: Elevation & Depth and Components — both stack-dependent and the implementing team's responsibility.

File template. Use this exact structure (replace bracketed placeholders with workshop output):

---
version: alpha
name: [Brand Name]
description: [One-line summary of the brand's visual identity]
colors:
  primary: "#XXXXXX"      # brand hero color — drives CTAs, headlines, emphasis
  secondary: "#XXXXXX"    # supporting brand color
  tertiary: "#XXXXXX"     # accent / highlight, NOT the hero
  neutral: "#XXXXXX"      # background / surface foundation
  surface: "#FFFFFF"      # default page background — override neutral if needed
  on-surface: "#XXXXXX"   # default text color on surface
  error: "#XXXXXX"        # semantic error, harmonized with palette
typography:
  h1:
    fontFamily: [Display Family]
    fontSize: 48px
    fontWeight: 600
    lineHeight: 1.1
    letterSpacing: -0.02em
  h2:
    fontFamily: [Display Family]
    fontSize: 32px
    fontWeight: 600
    lineHeight: 1.2
  body-md:
    fontFamily: [Body Family]
    fontSize: 16px
    fontWeight: 400
    lineHeight: 1.6
  label-sm:
    fontFamily: [Body Family]
    fontSize: 12px
    fontWeight: 500
    lineHeight: 1
    letterSpacing: 0.05em
rounded:
  none: 0
  sm: 4px
  md: 8px
  lg: 16px
  full: 9999px
spacing:
  xs: 4px
  sm: 8px
  md: 16px
  lg: 24px
  xl: 32px
  "2xl": 48px
  "3xl": 64px
---

# [Brand Name] — DESIGN.md

## Overview

A holistic description of the brand's look and feel: personality, target audience, the emotional response the UI should evoke. Whether it should feel playful or professional, dense or spacious. This is foundational context for an agent making stylistic decisions when no specific token applies.

## Colors

[Describe the palette's rooting principle — e.g., "high-contrast neutrals and a single evocative accent color".]

- **Primary (#XXXXXX):** [Descriptive name + usage guidance]
- **Secondary (#XXXXXX):** [Usage]
- **Tertiary (#XXXXXX):** [Usage — clarify this is for accents, NOT the hero]
- **Neutral (#XXXXXX):** [Foundation usage]

## Typography

[Describe the typography strategy — e.g., "Two distinct weights of Public Sans for narrative; Space Grotesk for technical labels."]

- **Headlines (h1, h2):** [Family, weight, role]
- **Body (body-md):** [Family, weight, role]
- **Labels (label-sm):** [Family, weight, role]

## Layout

[Describe layout strategy at a high level — e.g., "Fluid grid on mobile, fixed-max-width 1200px on desktop. 8px spacing scale with a 4px half-step for micro-adjustments. Components grouped via containment with 24px internal padding."]

## Shapes

[Describe corner-radius philosophy — e.g., "Architectural sharpness: 4px on interactive elements, 8px on cards, 16px on hero containers. Avoid mixing radii within a view."]

## Do's and Don'ts

- Do use `primary` only for the single most important action per screen
- Do maintain WCAG AA contrast ratios (4.5:1 normal text, 3:1 large text)
- Don't mix rounded and sharp corners in the same view
- Don't use more than two font weights on a single screen
- [Add 2–4 brand-specific guardrails surfaced during the workshop]

## Voice

3–5 bullet rules from the copywriter (e.g., "confident but not arrogant", "plain language over jargon", "contractions OK"). Custom section appended after Do's and Don'ts; preserved by spec-compliant consumers per the DESIGN.md spec's "unknown section" rule.

Validation before shipping. Run these checks against the emitted file:

# 1. File exists at output root, named exactly DESIGN.md (uppercase)
test -f <brand-root>/DESIGN.md && echo "DESIGN.md present"

# 2. YAML front matter starts the file
head -1 <brand-root>/DESIGN.md | grep -qE '^---$' && echo "YAML front matter starts correctly"

# 3. Required sections present in canonical order
grep -nE "^## (Overview|Colors|Typography|Layout|Shapes|Do's and Don'ts|Voice)" <brand-root>/DESIGN.md
# Should output 7 lines in spec order

# 4. (Optional) Lint with Google's CLI if available — flags broken refs, contrast issues, orphaned tokens
npx @google/design.md lint <brand-root>/DESIGN.md

Quality rule. If a section has nothing substantive to say, drop the section rather than write placeholder text. The spec preserves omitted sections without erroring. An honest "tokens only" file beats a bloated fake one.

Pitch Deck — Out of Scope

Brand-workshop does not emit a pitch-deck template. The pitch-deck skill reads DESIGN.md directly and generates its own brand-skinned Reveal.js deck — there is no intermediate template artifact for it to consume. Pre-emitting a deck here would duplicate work and fork the styling source of truth.

If the founder wants a deck, recommend they invoke pitch-deck (or team-composer with @startup_strategist + @vc_partner) after this workshop completes. pitch-deck will read the DESIGN.md this skill produced.

Output Files

Save all outputs to the resolved brand folder (see Phase 1 Step 0.0), organized into subfolders so the deliverable reads like a launch-day kit:

<brand-root>/
├── brand-brief.md
├── descriptions.md
├── DESIGN.md
├── logos/
│   ├── logo.svg
│   ├── logo-64.png
│   ├── logo-256.png
│   └── logo-512.png
├── favicons/
│   ├── favicon.svg
│   ├── favicon-16.png
│   ├── favicon-32.png
│   ├── favicon-180.png
│   ├── favicon-512.png
│   ├── site.webmanifest
│   └── favicon-install.html
├── mascot/                      (only when the mascot lane runs)
│   ├── mascot.md                (both paths — the recipe is the master)
│   ├── prompt-brief.md          (Path B only — model-agnostic generation brief; no generator reached)
│   ├── master.png               (Path A only — a generator ran)
│   ├── master-transparent.png   (Path A only)
│   ├── poses/                   (Path A only)
│   ├── character-sheet.html     (Path A only)
│   └── character-sheet-print.html (Path A only)
└── social/
    ├── og-image.png
    ├── x-header.png
    ├── linkedin-banner.png
    ├── instagram-square.png
    └── profile-avatar.png

No PDFs appear in the map: pagegate PDFs are verification-only and are deleted after the gate.

Where <brand-root> resolves per Phase 1 Step 0.0:

• docs/startup-kit/brand/ — orchestrated (via startup-launch-kit)
• docs/brand/ — solo default
• docs/startup-kit/brand/ — solo with docs/startup-kit/ smart default
• ${STARTUP_KIT_DOCS_ROOT}/brand/ — env-var override

Minimum viable set: If time or tooling is constrained, ship in this order of priority: brand-brief + logo → descriptions → favicons → DESIGN.md → social banners. The mascot lane is optional and always last when it was not explicitly requested. When the mascot or character sheet is the user's ask (e.g. "design a mascot and character sheet for my brand"), it moves into the minimum viable set — never drop the only deliverable the user came for.

Present all files to the user using present_files.

Edge Cases

• User only provides a name and one sentence: Run a minimal workshop (3 roles). Make assumptions explicit in the brief. Ask 1-2 clarifying questions max before proceeding.
• User has existing brand colors/fonts: Respect them. The visual designer works within those constraints and notes it.
• User wants multiple logo options: Generate 2-3 SVG variants and present them for selection.
• User wants tagline in multiple languages: Generate the primary in English, then produce culturally adapted (not literal) translations. Note: use the i18n skill if available for quality translations.
• Non-Latin scripts in logo: Flag that SVG text rendering for CJK/Thai/Arabic may need font embedding. Recommend logotype as a separate asset for those scripts.

Iteration

After presenting the first output, the user will likely want refinements. Common requests:

• "Make it more playful / serious / minimal / bold"
• "Try a different color palette"
• "I prefer tagline #3 instead"
• "Can you make the negative space more obvious?"

For these, re-run only the affected phase (don't redo the full workshop unless the direction fundamentally changes). Update the brief to reflect changes.

Closing — suggest the next step (light gate)

After the brand kit is presented, end the response with a one-line suggestion to run validation-canvas next:

"Brand without a validated market is a logo on a hypothesis. Next step: run validation-canvas to articulate what you believe about the problem, segment, and economics. The canvas reads the DESIGN.md tokens automatically from the resolved brand folder."

This is a light gate — informational, not enforced. The founder is free to ship just the brand kit; this skill's job ends with the kit. But surfacing the next step prevents the common failure mode where a founder ships a beautiful identity for an idea they haven't pressure-tested.

Quality Checklist

Before presenting final output, verify:

Core brand

• Tagline is concise (≤8 words ideal), memorable, and relevant
• Logo works at small sizes (mentally test at 32px)
• Color palette has sufficient contrast (check primary on white and dark backgrounds)
• SVG is valid and renders correctly
• Brief captures the reasoning, not just the output
• No cultural red flags were ignored

Extended deliverables

• Favicon is legible at 16×16 — no illegible detail; simplified variant used if needed
• site.webmanifest uses the actual brand theme_color (palette primary)
• favicon-install.html contains valid, copy-paste-ready <link> tags
• Social banners have ≥8% edge padding on the shorter axis
• Tagline appears only on hero banners (OG, IG square); profile headers stay clean
• All PNG banner files match their declared dimensions exactly
• Descriptions pack: every variant is under its hard character limit (verified, not estimated)
• Descriptions pack: each variant stands alone — no truncation chains
• <brand-root>/DESIGN.md exists as a standalone file — NOT folded into brand-brief.md as a section. Filename is exactly DESIGN.md (uppercase) per the Google Labs spec. Verify: test -f <brand-root>/DESIGN.md && echo OK.
• DESIGN.md begins with valid YAML front matter (delimited by --- lines). Verify: head -1 <brand-root>/DESIGN.md returns exactly ---.
• YAML front matter contains the cross-plugin contract key colors.primary. Verify: grep -E '^ primary:\s*"#' <brand-root>/DESIGN.md returns one line. Downstream plugins (validation-canvas, riskiest-assumption-test, pitch-deck) bind their accent CSS variable to this token.
• DESIGN.md sections appear in canonical spec order: Overview → Colors → Typography → Layout → Shapes → Do's and Don'ts → Voice. Verify: grep -nE "^## (Overview|Colors|Typography|Layout|Shapes|Do's and Don'ts|Voice)" <brand-root>/DESIGN.md shows the seven headings in order.
• Design system stays within tokens — no button/form/grid specs (Components and Elevation sections are intentionally omitted at "starter" scope).
• Empty DESIGN.md sections are dropped rather than filled with placeholder text. The spec preserves omitted sections without erroring.
• No deck/ subfolder is emitted under <brand-root>/. Brand-workshop does not pre-build a pitch-deck template — pitch-deck reads DESIGN.md directly. Verify: [ ! -d <brand-root>/deck ] && echo OK.

Mascot lane — generated assets (Path A: a generator ran)

• Template integrity: python3 <skill-dir>/scripts/verify.py integrity <brand-root>/mascot/<sheet>.html PASS on both instantiated sheets — zero leftover {{, balanced CSS braces, tags parse, every var(--x) defined, alt text on every image (see the command-convention note above for <skill-dir> / <brand-root>)
• Palette anchors: python3 <skill-dir>/scripts/verify.py anchors <brand-root>/mascot/master.png NAME=#RRGGBB ... PASS — every declared anchor ≥ 0.5% pixel share (tol 60)
• Print fit: python3 <skill-dir>/scripts/verify.py pagegate <brand-root>/mascot/character-sheet-print.html --landscape = exactly 1 page on Letter AND A4, engine Chromium (SKIPPED is not a pass — install playwright + pypdf)
• Cutout report sane — read the printed report, not the exit code: ≥1 kept component, pale (shadow) components dropped, coverage 30–60% for a full-body chibi
• Pagegate PDFs deleted after the gate — they never ship

Mascot lane — prompt brief (Path B: no generator reachable)

• Shipped <brand-root>/mascot/prompt-brief.md + <brand-root>/mascot/mascot.md (recipe skeleton) — the generated-asset gates above do NOT apply; never fabricate PNGs/sheets or report a gate that did not run
• Did not downgrade to a code-drawn mascot

Mascot lane — either path

• No lane bleed: logo/favicon lanes untouched; favicon does not derive from the mascot raster

**Shippin

…(truncated)