Brandprint

Brandprint

Purpose

Kill the generic-AI look (grey boxes, purple accents, stock fonts) on every output this repo produces. Any artefact a human will see — for Synthex itself, a portfolio brand, or a client — carries the correct brand automatically. The user should never have to say "make it branded".

Unlike a single-brand brandprint, this skill holds no brand data of its own. The single source of truth is packages/brand-config (typed .ts + agent-readable .design.md per brand). This skill teaches you to resolve, read, and apply it.

Step 1 — Resolve the brand

Exactly one owning brand per output. Resolution order (first match wins):

1. Explicit mention — the request names a brand, client, or organisation.
2. Organisation context — org-scoped work (an org's report, invoice, deliverable) uses that organisation's brand.
3. Project context — work inside a portfolio project uses that project's brand (e.g. Disaster Recovery work → dr).
4. Default — synthex.

Known slugs live in the BrandSlug union in packages/brand-config/src/types.ts and the registry in packages/brand-config/src/brands/index.ts. Current: synthex, dr, nrpg, ra, carsi, unite, john-coutis.

Unknown brand ⇒ STOP. If the resolved brand has no brand-config entry, do not improvise a palette. Either ask which existing brand applies, or onboard the brand properly via references/audit-recipe.md then references/build-recipe.md.

Co-brand mode — the only way two brands share an artefact

Two brands in one output is a separate mode, never a blend. Enter it only when all three preconditions hold; if any is missing, ask. Never invent a hybrid to resolve ambiguity.

1. An explicit request. The user asks for a co-branded artefact and both brands are identifiable. An implied pairing — "put the client's logo on our report somewhere" — is not a request for co-branding. Ask.
2. A resolved owner. One brand is the owner and the other the guest. The owner is whichever brand Step 1's resolution order returns: a deliverable produced for a client is owned by that client; Synthex's own collateral that credits a partner is owned by synthex. If the order returns both or neither cleanly, ownership is ambiguous ⇒ ask.
3. Both brands in brand-config. Unknown brand ⇒ STOP applies to the guest as well. No brand-config entry, no guest slot — onboard it first.

Once those hold, the split is fixed. It is not renegotiated per artefact:

The guest's entire presence is its logo plus its doNot list. Nothing else crosses the line: no guest accent colour "for balance", no guest display font on the headings, no guest tone of voice.

Lock-up rules

• Order: owner's mark first — left of the guest horizontally, above it vertically. Right-to-left locales mirror the axis, not the precedence.
• Separation: a hairline divider or plain whitespace between the marks. Never overlap, interlock, or merge them, and never set them on a shared coloured plate belonging to neither brand.
• Clear space: the larger of the two brands' safeAreaPx, applied both around the whole lock-up and between the two marks. When one brand's rules are stricter, they govern the lock-up.
• Variants: choose primary vs inverted per brand against the actual background, following each brand's own logo rules. The two variants need not match.
• Scale: optically match the marks' visual weight. Never stretch, crop, recolour, or regenerate either mark to make it fit — logos are files (see Hard boundaries).
• Placement: one lock-up per artefact, in the masthead or the footer. Not repeated per page, per slide, or per section.

Anything the two brands' rules cannot settle between them is [MISSING] — ask, per the evidence discipline in Step 2.

Step 2 — Read the brandprint

Load both files for the resolved <slug>:

When the brand is synthex, read .claude/DESIGN.md as well — it is a third source, not a substitute for either file above. Its own header states it is the agent-readable projection of packages/brand-config/src/brands/synthex.ts plus the founder's non-negotiable rules, and it records known divergences elsewhere in the repo. So Synthex takes three files; every other brand takes two. Cite whichever you actually read (see Verification).

Evidence discipline (per .claude/rules/fabel-evidence-standard.md): values read from brand-config are [VERIFIED]. Anything you deduce beyond it (a tint, a pairing, a layout choice the spec doesn't cover) is [INFERRED] and must be flagged in your summary. A value the brand needs but the config lacks is [MISSING] — ask, never invent. No made-up hex codes, no substituted fonts, no "close enough" logos.

Step 3 — Apply per surface

Copy in any surface obeys the brand's voice block: tones, forbiddenWords, requiredCadence, and the doNot list. For strategic voice decisions defer to brand-strategist; mechanical gating is brand-voice-enforce.

Web/UI surfaces — hand off to impeccable

For web pages, app UI, and components, apply the brand through the impeccable skill (installed at ~/.claude/skills/impeccable/). The brand's <slug>.design.md is already in the Google DESIGN.md format impeccable consumes — point it at that file (or .claude/DESIGN.md for Synthex) so it inherits the brand's tokens instead of inventing its own. Impeccable's absolute bans (gradient text, AI-beige, eyebrow kickers, identical card grids, side-stripe borders) are the anti-slop gate for branded UI.

Hard boundaries

• Logos are files, never generated. Use only repo asset paths: public/logos/<slug>/{primary,inverted,icon}.svg, public/synthex-logo.*, or the paths in the brand's logo block. Generating, redrawing, or approximating a logo violates the Real Images Only rule (.claude/rules/real-images-only.md). public/brands/<name>/ is a legacy scheme keyed by full brand name — read from it where existing code already does, never write new brand assets there (SYN-1133).
• No brand's declared logo files exist yet. All 21 paths in the seven brands' logo blocks are absent from disk, tracked in brand-logo-baseline.json (SYN-1133); public/logos/ holds neither scheme today. Nothing reads the field at runtime, which is why the gap never threw. So for any brand but synthex — whose marks are at public/synthex-logo.* — a logo is [MISSING]: say so and ask for the artwork. Do not substitute another brand's mark, a flat customer logo, or a generated one.
• All imagery defers to grounded-visuals — owned reference library via lib/services/ai/image-generation.ts, no exceptions.
• One brand owns each output. Never blend palettes, fonts, or logos across brands, including "Synthex plus client" hybrids. A genuine co-branded artefact is not a blend: it runs Co-brand mode, where the owner supplies every token and the guest appears only in the lock-up. No explicit request, or no clear owner ⇒ ask.
• Australian English in all product copy: colour, organise, recognise, licence (noun), authorise.

Onboarding a new client brand

Three steps, in order — skipping the audit is how brands come out "a little bit off":

1. Audit — references/audit-recipe.md: inventory the client's assets, extract the spec, tag every value [VERIFIED] / [INFERRED] / [MISSING].
2. Resolve flags — the founder/client answers the audit's question list. Answering those questions is the whole game; do not build on unconfirmed values.
3. Build — references/build-recipe.md: turn the corrected audit into a first-class brand-config entry, then verify with the real commands.

Verification

Before claiming a branded output done:

• Name the resolved brand slug and cite every file you actually read, by path — not a count. That is <slug>.design.md + <slug>.ts for any brand, plus .claude/DESIGN.md when the brand is synthex (three files), plus the guest's two files in co-brand mode. A file you read and did not cite is an untraceable claim.

• Confirm every colour/font/logo traces to brand-config ([VERIFIED]) or is flagged.

• Never restate token values in this skill or any other document as though they were the source. Quote them from brand-config at the point of use, so a token correction lands in one place.

• Co-brand mode also requires: naming which brand is the owner and which is the guest and why; confirming the guest appears only in the lock-up; stating the safeAreaPx used and which brand it came from; and listing the union of both doNot lists you checked against.

• For HTML/CSS outputs, optionally run the deterministic slop gate — the detector bundled with the installed impeccable skill, over local files:

  node ~/.claude/skills/impeccable/scripts/detect.mjs --json <files>
  

  Never npx impeccable. impeccable is not a dependency of this repo — it is absent from package.json and node_modules/.bin — so npx would resolve the bare name against the public registry and execute whatever it found there. The bundled detector is version-pinned by installation (v3.9.1, pbakaus/impeccable @ f2049c2, per the source line at the foot of ~/.claude/skills/impeccable/SKILL.md), reads local files with no network, prints JSON, and exits 2 when it finds anything and 0 when clean. Its rules live in ~/.claude/skills/impeccable/scripts/detector/registry/antipatterns.mjs.

• Banned: "should look right", "probably on brand". Show the values used.