brand-generator

Brand Generator

Reads brand/brief.md and produces concrete brand assets in the brand/ directory.

Preconditions

• brand/brief.md must exist with both Part 1 (user answers) and Part 2 (inferred direction). If it doesn't, invoke the brief-builder skill first. Never fabricate a brief.
• Python 3.9+ available (the rasterize and OG scripts need it).
• Dependencies in scripts/requirements.txt installed. If not installed, run pip install -r <path-to-this-skill>/scripts/requirements.txt once. Tell the user before installing.

What to generate

For a full kit (/stamp:init), produce all of these. For /stamp:finish hero, only run the OG composition step.

1. brand/voice.md

A short voice and tone guide derived from Part 2's voice adjectives, audience tier, and tone constraints. Sections:

• Adjectives — the 3–5 from the brief, each with a one-line "this means…" gloss.
• Do / Don't — concrete writing rules. Pull "Don't" items directly from the user's tone-constraints answer.
• Sample sentences — 3 example product copy sentences (e.g., a hero headline, a button label, an error message) in this voice. Make them about this project, not generic.

Keep this under one page. Long voice guides don't get read.

2. brand/palette.md

A palette with 4–6 colors, picked using a recipe from references/palette_recipes.md. Use the palette family named in Part 2 of the brief.

For each color, list:

• Role (e.g., "primary", "ink", "surface", "accent", "warning").
• Hex value.
• One-line usage note.

After the swatches, include a "Contrast pairs" subsection listing which foreground/background combinations meet WCAG AA (4.5:1 for body, 3:1 for large text). This is non-negotiable — if the brief specified an accessibility floor, the palette must satisfy it. Adjust hex values until it does.

3. brand/fonts.md

A type pairing chosen from references/font_pairings.md. Use the family named in Part 2.

Include:

• Display / heading face (name, weight, source — Google Fonts URL or system).
• Body face (name, weight, source).
• One-line rationale tying the choice to the brief.
• Two-line CSS @import or <link> snippet the user can paste.
• Recommended size scale (e.g., 48 / 32 / 24 / 18 / 16 / 14).

4. brand/logo.svg

A primary logo concept following a style from references/logo_styles.md (use the style named in Part 2).

Constraints:

• Pure SVG, no embedded raster.
• Single color path where possible (so it can be inverted for dark backgrounds without re-export).
• 1024x1024 viewBox is the default; adjust if the style requires landscape.
• The wordmark, if present, should use the brand display face — but reference it by font-family and also include a fallback path version (in case the user views the SVG without the font installed). For v0.1, you may prioritize the symbol mark over a wordmark to avoid font-embedding complexity.

After writing the SVG, run scripts/rasterize_logo.py to produce PNGs at sizes 16, 32, 64, 128, 256, 512, 1024 in brand/logo/.

5. brand/og/og-default.png

A 1200x630 OG image with: brand background color, the rasterized logo, the project name in the brand display face, and a one-line tagline drawn from the brief's function answer.

Run scripts/make_og.py with no --hero arg to produce this. Output goes to brand/og/og-default.png.

How to choose recipes

Part 2 of the brief names a palette family, type pairing, and logo style by name. Do not second-guess those choices — the brief-builder picked them based on the user's actual answers, and changing them here would silently undo that work. If you genuinely think a different recipe fits better, ask the user rather than swap.

If Part 2 is missing one of those names (older brief, manual edit), pick the most defensible choice from the reference file matching the audience tier and visual register, and note in the generated file which recipe you used and why.

How to call the scripts

The scripts live in <this-skill>/scripts/. Resolve their absolute path from the skill's location — do not hardcode ~/.claude/....

python <skill-path>/scripts/rasterize_logo.py \
  --input brand/logo.svg \
  --output-dir brand/logo

python <skill-path>/scripts/make_og.py \
  --brief brand/brief.md \
  --palette brand/palette.md \
  --fonts brand/fonts.md \
  --logo brand/logo/logo-512.png \
  --output brand/og/og-default.png

For /stamp:finish hero <path>, add --hero <path> to the make_og call and change the output to og-hero.png.

If a script fails (e.g., cairosvg missing), surface the actual error to the user with the install command they need, then stop. Don't fall back to a less-good output silently.

When to overwrite vs preserve

• If a target file already exists, ask before overwriting, except:
  • Rasterized logo PNGs in brand/logo/ are derived from brand/logo.svg — always regenerate when the SVG changes.
  • brand/og/og-default.png is also derived; regenerate freely.
• Voice, palette, fonts, and the logo SVG are creative outputs — confirm before overwriting.

Final report

After generation, print to the user:

• A bulleted list of every file written, with its path.
• The hex codes for the primary and accent colors (so they can eyeball them in the terminal).
• The next-step suggestion: /stamp:finish hero <path> if they have a hero image.