Train a Brand Design Guidelines Skill

Objective

Learn the visual system behind my brand, document which rules are authoritative versus inferred, and package those rules so an AI can create and review new assets without drifting into generic design.

Evidence boundaries

1. 1

   Treat an official brand guide as the highest-authority source, followed by approved design-system files, then approved examples.

2. 2

   Never treat an arbitrary old asset as a rule just because it exists. Flag conflicts and ask which source wins.

3. 3

   Do not redraw, distort, recolor, or invent a logo. Do not fabricate colors, fonts, licenses, partner marks, photography rights, or accessibility claims.

4. 4

   Preserve original source files. Work from copies or references and do not modify approved assets.

5. 5

   Label every guideline as Documented, Observed, or Proposed:

   • Documented: stated in an authoritative source.
   • Observed: repeated consistently across approved examples.
   • Proposed: a gap-filling recommendation that requires my approval.
6. 6

   Do not create the final skill until I approve the evidence audit and proposed system.

Audit the source material

Inventory every supplied source with

• Source name and direct file or document link
• Artifact type and date, when known
• Authority level
• What it contributes to the system
• Conflicts, gaps, or usage restrictions

Analyze only examples I identify as approved. When examples disagree, preserve the disagreement instead of averaging it into a new rule.

Extract the visual system

Identity and logo

• Approved logo variants and when each is used
• Clear space, minimum size, alignment, and background rules
• Co-branding or partner-lockup rules
• Explicit incorrect-use examples

Color

• Primary, secondary, accent, neutral, background, and semantic roles
• Exact supplied values for digital and print; never estimate a color by eye when a specification exists
• Permitted combinations and contrast requirements

Typography

• Approved font families, fallbacks, weights, and licenses or availability
• Type roles for display, heading, body, label, caption, and data
• Size, line-height, tracking, case, and emphasis patterns by artifact type

Composition

• Grid, margins, alignment, density, whitespace, spacing scale, border radius, stroke, and shadow patterns
• Repeated hierarchy rules that make an artifact recognizably ours
• Rules that change between web, slides, documents, and social graphics

Imagery and graphics

• Photography subject, composition, lighting, crop, and treatment
• Illustration, icon, texture, diagram, and data-visualization style
• What to avoid, with evidence from the approved sources

Accessibility and production

• Contrast, legibility, minimum type, alternative text, motion, and color-only communication rules
• Export sizes, file formats, bleed, safe zones, and responsive behavior when documented

Turn observations into usable guidance

For every rule, record

• The rule in plain language
• Status: Documented, Observed, or Proposed
• Source link and page, frame, or example location
• Where it applies
• One compliant example and one non-compliant example
• Confidence: high, medium, or low

Keep brand voice out of this skill except where short visual copy affects typography or layout. Refer copywriting work to the separate brand voice skill.

Define artifact modes

Create specific guidance for each requested mode rather than one universal style:

1. 1

   Website and product UI

2. 2

   Presentation slides

3. 3

   Documents and reports

4. 4

   Social and campaign graphics

For each included mode, define its canvas or breakpoints, hierarchy, grid, component patterns, image treatment, density, and final export checks. Omit unsupported modes instead of inventing rules.

Deliverables and approval gate

First, show me

• The source inventory and authority order
• The proposed design tokens and visual rules
• Conflicts and missing decisions
• Coverage by artifact mode
• The exact files and approved assets you plan to package

After I approve, create

brand-design-guidelines/ ├── SKILL.md ├── assets/ │ └── [approved logos, fonts, templates, or linked asset index only] └── references/ ├── brand-system.md ├── artifact-rules.md └── qa-checklist.md

Keep SKILL.md concise. It should explain when the skill applies, how to choose an artifact mode, which reference files to read, and which approval or asset restrictions must never be bypassed.

Put the evidence-backed tokens and identity rules in brand-system.md. Put mode-specific layout and production rules in artifact-rules.md. Put the preflight review in qa-checklist.md.

Required QA behavior

Before declaring any new artifact complete, the finished skill must:

1. 1

   Identify the artifact mode and intended audience.

2. 2

   List the approved assets and rules it is applying.

3. 3

   Check logo use, color, typography, hierarchy, spacing, imagery, accessibility, and export requirements.

4. 4

   Flag any unsupported design decision as proposed instead of silently inventing it.

5. 5

   Return a short compliance report with direct links to the supporting guidelines.

6. 6

   Require human review before publishing, printing, or replacing an approved template.

Validation

Test the draft skill against at least one representative artifact in every supported mode. Compare the result with an approved example, record each mismatch, revise the rule that caused it, and rerun the check. A visually polished output is not automatically on brand; the evidence and QA checklist decide.