Visual identity spec that makes on-brand unambiguous

STAFF RIDD / LISA · V0.2.0

gemskills:style-creator

REQUIRES

claude plugin install gemskills@b-open-io

WHEN TO REACH FOR THIS

"On-brand" is a matter of taste until it is written down - and taste does not survive being handed to an agent that generates a hundred assets. This is the document every other design playbook reads first: it fixes logo usage, color tokens, type scale, imagery rules, and - critically - a reusable style reference tile so that every future generation looks like it came from the same studio. The 2026 failure mode for AI design arrives after the logo exists, when every later asset drifts a little further from it. This spec is the anchor every later playbook generates against.

REQUIRES

gemskills plugin: claude plugin install gemskills@b-open-io
GEMINI_API_KEY - generating the reference style tile
Existing brand assets to reference (logo files, an existing site, or a moodboard) if you have them; the spec can also define an identity from scratch

CONTEXT INPUTS

{{BRAND_NAME}}       The name exactly as it should always be written (capitalization matters).
{{POSITIONING}}      One sentence: what you are and who it is for. Anchors the visual mood.
{{LOGO_FILES}}       Paths to logo lockups you have (primary, mark-only, mono). "none" if defining fresh.
{{COLOR_SEED}}       Your known colors as hex, or a description ("deep maroon + amber accents"). The primary/accent pair.
{{TYPE_PREF}}        Type direction: named fonts if fixed, else a feel ("geometric sans headlines, humanist body").
{{IMAGERY_MOOD}}     How pictures should feel: subject matter, palette, texture, what to avoid (e.g. "no stock-photo handshakes").
{{STYLE_ID}}         A short id for the reusable style you are locking, e.g. "acme-house". Lowercased, hyphenated.
{{OUTPUT_PATH}}      Where the spec doc lands, e.g. ./brand/VISUAL-IDENTITY.md.

THE PROMPT

PASTE INTO YOUR AGENT

You are producing a Visual Identity Spec - the source of truth every later design
task generates against. Ridd (UI Designer) owns the structured rules; Lisa (Visual
Content Specialist) owns the reusable style reference.

Write {{OUTPUT_PATH}} as a markdown spec with these sections, each stated as rules
an agent can obey without judgment calls:

1. Name & lockups - {{BRAND_NAME}} written correctly; when to use each lockup from
   {{LOGO_FILES}}; clear-space and minimum-size rules; what NOT to do to the logo.
2. Color tokens - from {{COLOR_SEED}}, define named tokens (background, surface,
   primary, accent, text, muted) as hex, each with a plain-language "use for" note.
   Give the light and dark value for every token; never a single hardcoded color.
3. Type scale - from {{TYPE_PREF}}, name the heading/body/mono families and a
   concrete size + weight scale (display, h1-h3, body, small, code).
4. Imagery rules - from {{IMAGERY_MOOD}}, state subject, palette, texture, and an
   explicit avoid-list so generated images stay in one family.
5. Reusable style reference - the anchor. Load `gemskills:style-creator` and follow
   it to define a new style with id {{STYLE_ID}}: write promptHints (8-12 palette/
   texture/mood descriptors, NOT scene objects) and a tilePrompt built with the
   skill's 6-part structure (full-bleed anchor, macro/texture anchor, style
   descriptors, materiality, palette/lighting, anti-container enforcers). Add the
   entry to the style library's styles.json and generate its 512x512 tile with the
   skill's generate_tiles script. This is what {{IMAGERY_MOOD}} becomes mechanically.
   Record in the spec: "generate all brand imagery with --style {{STYLE_ID}}".

Ground every rule in the inputs - do not invent colors or fonts I did not give.
Where an input is "none" or vague, propose a value and mark it PROPOSED so I can
confirm rather than silently guessing.

Do NOT read the generated tile back into context (style-creator outputs a path);
ask me to visually inspect it for full-bleed texture transfer (no frames, no
scene, right palette) and iterate the tilePrompt if it fails.

Negative prompts: no single-value colors (every token gets light + dark); do not
describe imagery as a scene in the tilePrompt (transfer materiality, not a picture);
do not skip the anti-container enforcers; do not leave any section as prose I still
have to interpret - it must be rules an agent can follow verbatim. Do not run
style-creator's maintainer steps (updating style counts, bumping the gemskills
plugin version, or committing to the plugin); stop once the styles.json entry is
added and the tile is generated.

EXPECTED OUTPUT

{{OUTPUT_PATH}} - a complete visual identity spec with name/lockup rules, light+dark color tokens, a concrete type scale, imagery rules, and a locked reusable style id - plus that style registered in the library with its reference tile generated. Every later design playbook can now cite --style {{STYLE_ID}} and inherit the same look.

FOLLOW-UP CHAIN

db-brand-voice-sheet - the verbal half of this identity, so copy matches the visuals.
db-design-token-manifest - turn the color tokens here into the code-side shadcn source of truth.
db-logo-concepts - if {{LOGO_FILES}} was "none", generate directions against this spec.
db-brand-identity-system - roll the whole identity out into a delivered asset kit.

← PACK CONTENTS