lucky-branding

lucky-branding — disciplined five-phase brand identity

Use when building or refreshing a brand identity. End state: a complete brand package handed

off as production-ready files + a machine-readable guidelines.md + voice.md + a tokens.json

that passes the handoff gate. **Five sequential phases with an iterative feedback loop. The

output of phase N feeds phase N+1. No skipping ahead.**

The methodology lives in workflow.md

The full phase-by-phase methodology — exact document structures, deliverables, and approval gates — is in [workflow.md](workflow.md). That file is the source of truth. This skill is the enforcement layer: when to use the workflow, what counts as "approved" before moving on, and the canonical failure modes that come from shortcuts.

Read workflow.md once at the start. Re-read between phases. Do not paraphrase its structure here — link to it. If the skill and the workflow doc drift, the workflow doc wins.

The token architecture (the part most brand processes skip) is in [tokens/schema.md](tokens/schema.md), with the shipping gate in [tokens/phase4-handoff-gate.md](tokens/phase4-handoff-gate.md) and a working Style Dictionary build in [tokens/](tokens/).

When to use

• •Starting a brand identity for a new product or company
• •Refreshing an existing brand that has partial assets but no structured package
• •Producing deliverables an outside designer or freelance studio could pick up and continue from

When NOT to use

• •One-off marketing collateral (a single banner, a single deck) — overkill
• •A single color change to an existing brand — edit guidelines.md directly
• •Pure voice/copy guidance without visual identity — write only voice.md
• •Just a logo — use the Phase 2 portion only, not the full skill

Inputs

• •brand_name — slug, lowercase-hyphen
• •target_root — brands/<brand_name>/
• •Initial project brief — at minimum: what the brand does, who buys it, why it exists

Outputs (chained — each phase's output IS the next phase's input)

What this skill enforces beyond the workflow doc

The exact contents of each phase are in workflow.md. Below is the discipline that prevents shortcuts — read it as a set of tripwires.

Phase 1 — Creative Direction

• •Use the exact 8-section structure from workflow.md Phase 1.
• •Two aesthetics, not four, not one. Two genuine alternatives across three axes each

(Logo & Graphics / Colour & Texture / Font & Voice) — never "the chosen one" + a strawman.

• •Iterate by version (-0.2.md), never overwrite — the version trail is the record.
• •Gate: client picks ONE aesthetic and answers the 4 questions before Phase 2.

Phase 2 — Logo Concept

• •Write logo-concept-0.1.md — the brief + render plan, including the render-tool decision.
• •Pick the right tool (per workflow.md Phase 2): Tool A (SVG, for typographic/geometric

marks) vs Tool B (image-gen, for illustrative/textured marks). Read the chosen Phase 1 Aesthetic to decide. Don't dress one as the other.

• •Output: brief + rendered files in a logo-concepts/ subfolder + an inspection HTML loading

all directions side by side for review.

• •Gate: client picks ONE direction before Phase 3.

Phase 3 — Brand Guidelines

• •All 8 sections from workflow.md Phase 3.
• •5 colors. 3 fonts. Canonical counts. Color codes are HEX + RGB + CMYK + Pantone for

each — not just HEX. That's what makes it print-ready.

Phase 4 — Brand Package + token gate

• •The 4-folder structure from workflow.md Phase 4.
• •`tokens.json` must pass `tokens/phase4-handoff-gate.md` before the package is "done":

semantic layer present, primitives separated, dark mode defined as a semantic re-map, the Style Dictionary build succeeds for all three platforms, typography tokenized, zero interpretation debt.

Phase 5 — Text Artifact Extraction

• •ONLY now produce guidelines.md + voice.md.
• •guidelines.md carries the full HEX+RGB+CMYK+Pantone set (system of record), even if

implementation uses HEX only.

Acceptance (full skill)

• •[ ] Phase 1: creative-direction-0.x.md has all 8 sections in order, 2 distinct aesthetics

(3 axes each), 4 next-step questions; client-approved (one aesthetic, 4 answers recorded)

• •[ ] Phase 2: logo-concept-0.x.md brief exists with the render-tool decision; marks rendered

with the right tool; primary + secondary + mini-mark, 3 colour variants each; client-approved

• •[ ] Phase 3: brand-guidelines.pdf (or .md) has all 8 sections, 5 colors with

HEX+RGB+CMYK+Pantone, 3 fonts with 3 roles

• •[ ] Phase 4: 4-folder structure with all production file formats; tokens.json passes the gate
• •[ ] Phase 5: guidelines.md + voice.md complete and machine-readable
• •[ ] Brand CONTEXT.md updated with status per phase

Anti-patterns

The canonical shortcuts that ruin a brand build. All are the kind of thing an AI does on auto-pilot when it senses "let's just ship." Treat each as a tripwire.

• •Skipped Phase 1; jumped straight to color tokens + Tailwind config. The #1 failure. No

written Phase 1 means no record of why the aesthetic was chosen, no version trail, no audit of alternatives. Restart at Phase 1; absorb the work already done into the retrofit.

• •Phase 1 with ONE aesthetic + a strawman. Two genuinely build-able alternatives is the

discipline. A strawman makes Phase 1 ceremonial and lets the AI commit to its first instinct.

• •Phase 1 with 4 aesthetics, not 2. The client can't seriously evaluate 4 directions in one

pass. If you're tempted by 4, the upstream narrowing wasn't done. Two.

• •Hand-coded SVG for an illustrative logo (animals, textured marks). Image-gen is the right

tool here; SVG produces stiff, wrong output.

• •Image-gen output for a typographic logo. Raster approximates letterforms badly. SVG with

the actual web font via embedded @import IS the production format — no vectorisation step.

• •Conflating the two tool modes. Read the chosen Aesthetic. Illustration/texture → image-gen.

Letterforms/geometry/typography → SVG.

• •Phase 3 with 3 colors and 1 font. Canonical counts: 5 colors, 3 fonts. If the brand

doesn't "need" 5, define the missing slots anyway — print-readiness depends on them.

• •`guidelines.md` with HEX only, no RGB/CMYK/Pantone. Phase 3 was abbreviated and Phase 5

inherited the shortage. Print-readiness fails. Add the missing codes to both.

• •Calling an HTML one-pager the deliverable. The HTML is a useful internal accelerator;

Phase 3's deliverable is the PDF. The PDF ships.

• •Shipping flat hex tokens with no semantic layer. The handoff gate exists to stop exactly

this. A tokens.json that names colors by appearance (green-500) instead of use (action-primary) pushes brand decisions onto the implementer. Fix before shipping.

• •Phase 5 produced first, Phases 1–4 skipped. The AI optimized for "make it work in code

fast." Code implementation IS Phase 5 — but only because 1–4 fed into it. Reset; produce the missing phases. The audit trail IS the brand.

• •Starting without reading `workflow.md`. This skill is enforcement; the workflow doc is the

source. Reading only the skill is reading the cover, not the book.

Why the order is non-negotiable

A brand assembled out of order — tokens first, creative direction never — works in code and fails everywhere else: no rationale, no version history, no defensible "why," nothing a fresh designer or a client can pick up and continue. The five phases force the thinking before the making, and the token gate forces the brand to be consumable mechanically at the end. That's the whole product: a brand that survives the handoff.