For AI agents

---
name: galton-brand
description: Routes any GALTON-branded output task (UI, pages, slides, posts, charts, copy, images) to the right knowledge via MCP tools or the read URLs. Use before creating any GALTON-branded asset.
compatibility: Requires network access to https://galton-ai-assets.vercel.app. MCP server at https://galton-ai-assets.vercel.app/api/mcp preferred; the read URLs work without MCP.
metadata:
  brand-version: "2.21.1"
  source: https://galton-ai-assets.vercel.app/skill.md
---

# GALTON Brand Layer, the orchestrator skill

You are working with the GALTON brand, served by a read-only Brand Layer knowledge base: MCP server at https://galton-ai-assets.vercel.app/api/mcp (preferred) or the read URLs. It returns brand data only: YOU produce the asset in your own tools.

No brand tools visible? Connect first: `claude mcp add --transport http galton-brand https://galton-ai-assets.vercel.app/api/mcp` (or your client's HTTP-MCP equivalent). No MCP at all? Use the read URLs below, or POST to https://galton-ai-assets.vercel.app/api/tool, which runs the same tools over plain HTTP. If you are an agent inside a browser, open any page of https://galton-ai-assets.vercel.app: it registers these tools with you directly over WebMCP, with nothing to configure.

## Keep this skill current

This skill is generated from the brand and carries its version (`2.21.1`). At the start of a session call `get_brand_overview` (it returns `version`; or read https://galton-ai-assets.vercel.app/brand.json meta) and compare versions. If they differ, re-fetch https://galton-ai-assets.vercel.app/skill.md and replace your saved copy. Saving the skill is optional: if your environment supports skills (e.g. `.claude/skills/galton-brand/SKILL.md`), save it there; otherwise just keep it in context.

## Hard rules (never break)

1. Never invent a colour, font, radius or component. The accent is ALWAYS #FFD700 (colors.primary), never red, blue or green. H1/display headlines and section titles (H2) are UPPERCASE. Every text colour is SOLID: no opacity, no alpha, no grey type, and no em dash in any copy. Anything undefined: derive via the fallback contract, or ask.
2. Call `get_agent_rules` once per session for the rules of engagement: the colour and type rules, the fallback protocol, the voice prohibitions and the starting prompts. It carries rules, not values; the values are in `list_tokens`, `get_fallback` and `get_voice_guide`.
3. Hotlink assets by their ABSOLUTE URLs (`list_assets`): never relativise, download, or retype a wordmark as text. ONE exception, and only this one: inside the unzipped offline package (https://galton-ai-assets.vercel.app/bundle.zip) the local path `./brand/X` **is** `https://galton-ai-assets.vercel.app/brand/X`, the same file shipped alongside this skill. On a machine with no network, use that path. Everywhere else the absolute URL stands.

## Guided mode: run this BEFORE you produce anything

Most requests are one sentence long and leave most parameters unset. Fill them from the defaults in the table below, ask only about what you cannot infer, report what you chose, then run the self-check.

The protocol itself is 9 numbered steps and it is not repeated here: read it once from `get_intake` (or https://galton-ai-assets.vercel.app/intake.md), which also carries the per-output parameters, the licence to ask your own question and its boundary, and the exact report block to append. This section is the summary you act from once you have.

Never ask more than 3 questions, and never ask twice. If the user replies `go` / `choď` / `chod`, produce immediately from the defaults.

| Output | Ask only about | Everything else defaults to |
|---|---|---|
| Poster / out-of-home (`poster`) | headline | surface: dark; format: 4:5 portrait; icons: none; image: no; logo: primary wordmark lockup; language: Slovak |
| Social post (visual + caption) (`social`) | topic | surface: dark; canvas: 1080x1350; layout: cover; icons: none; image: no; caption: yes; language: Slovak |
| Presentation / deck (`deck`) | topic | slides: 10; surface: mixed; ratio: 16:9; sequence: picked from the presentation catalog by the content: cover-main, agenda, a divider per chapter, the content slides, cover-closing; format: HTML; icons: line icons; chart: yes; language: Slovak |
| Table / data sheet (`table`) | data | size: 5 columns, 5 rows; surface: light; tags: no; language: Slovak |
| Chart / data visualization (`chart`) | data | type: bar chart; surface: dark; series: 1; language: Slovak |
| Copy / text (`copy`) | subject | register: consultative; length: short; cta: one soft next step; language: Slovak |
| Web page / UI (`web`) | purpose | surface: dark; sections: hero, three content sections, footer; components: yes; language: Slovak |

Then append the report block (`Brand choices`), in the user's language: every parameter you resolved, its value, and the one-word alternative. `get_intake` carries the worked example.

**Never a parameter, never offered as a choice:** The accent colour. It is always the brand primary, on every surface, in every state. The typeface and the type scale. Both come from the design spec. The logo. White or black only, never recoloured, never retyped as text. The UPPERCASE rule for H1/display headlines and section titles (H2). The chart palette. Charts never use a library-default rainbow. Status colours (success/warning/error/info). Functional feedback only, never decoration.

**Self-check before delivering:**
- The accent is the brand primary, used for emphasis only, never as a large fill.
- The case follows `typeRules.case`.
- The logo is white or black, hotlinked from its real asset URL, not redrawn or retyped.
- Every text colour came from a role or a `.bl-text-*` class and satisfies `colorRules.text`; on a light surface it came from the `lightSurface` map, not from a dark-tuned token.
- Every colour, size, radius and component came from the spec, or from `fallback` where the spec is silent.
- Any chart uses the chart palette; any icon comes from the icon library, in the variant that matches the surface.
- Copy follows the verbal identity (register, lexicon, mechanics), not generic marketing language, and contains no em dash.
- Anything the spec does not define went through the fallback contract.

## Route by task

| You are asked to make | Do this, in order |
|---|---|
| Anything (first step) | `get_brand_overview`: identity, surfaces, the text system, the critical rule. |
| Web / UI / HTML | Link the stylesheet `https://galton-ai-assets.vercel.app/brand.css` (fonts + token variables + `.bl-{component}` classes), take markup from `https://galton-ai-assets.vercel.app/snippets.json` or `get_component` (`snippetHtml`). Tokens: `list_tokens` (pass `ref` to resolve one). |
| Anything on a LIGHT surface | The brand is dark-first, so every token and every `.bl-{component}` class is the dark rendering. Do not invert it yourself: wrap the light area in `.bl-light` and keep the same classes (brand.css carries the overrides), or read the map with `list_tokens { category: 'lightSurface' }`. `get_component` returns the light `changes` for that one component. The accent never inverts, and a link on light is ink, not yellow. |
| A colour for type, anywhere | Never pick one yourself. On the web use the `.bl-text-*` classes; elsewhere take the role from `get_brand_overview.essentials.text`, which carries the rule with the values. |
| Copy / text of any kind | `get_voice_guide` FIRST: apply its `aiBrief`, registers and lexicon before writing a word. |
| LinkedIn post | `get_voice_guide` first, then `get_linkedin_playbook` (post types, anatomy, hashtag rules), then `list_carousel_recipes` for a carousel: it picks the slide sequence by post type. |
| Video, reel, presentation, event or demo with sound | `get_sound_guide`: pick the track by `jobs`, hotlink its `url`, follow `principles` (house tempo, instrumental under a narrator, the sound logo as the last sound). Never use stock or generated music in its place. |
| Which slide layouts exist | `list_layouts`, the fast index: every layout as one row with its example image and SVG source. Start here; it costs a fraction of the detail call. |
| Slides / carousel visuals | `get_layout { id }`, and add `ratio` for another canvas height: the server resolves the geometry for you, so never rescale a measured number by hand. You get that layout's slots, its MEASURED geometry in absolute px, its `must` list, a reference SVG, and the slice of the shared canvas it uses. Build the slide yourself from the numbers; never place an element by eye off the example image, and never drop the rule, the logo or the URL. |
| A slide question the layout does not answer | Call `get_layout` with no arguments once per session. That is where the rest of the canvas lives: the legends for `sizing` (before you resize any box), `slot` and `verified`, `ratioAdaptation`, the composition model, and `imageTreatment` (before you put type over a picture). It is identical in every reply, which is why a layout call carries only the slice it needs. |
| A presentation / deck (16:9) | Its own catalog, not the social layouts. PLAN with `list_presentation_slides`: pick every slide by what its content has to do (intent, density, limits) and obey the deck rules (cover first, closing last, the full yellow box at most once). Then `get_presentation_slide` without id ONCE for the 1920x1080 canvas (type scale, spacing, footer), and `get_presentation_slide { id }` per slide type for its slots, must list and measured boxes. Never stretch a slide past its item count: split the content over two slides. |
| A canvas the catalog does not measure | The measured canvases are the ones `get_layout` lists under `ratios`, plus the 1920x1080 deck canvas of the presentation catalog; every other size (a 4:3 deck, A3 print) has NO geometry, and `get_layout` refuses the ratio rather than guess. Build it from the shared canvas: the padding, the type scale, the chrome bands and the anchoring rules in `ratioAdaptation`. Then say in the brand-choices block that the canvas was derived, not measured. |
| Icons | `list_icons`. Without args you get the whole library, one row per icon with its hotlinkable `url` (dark surfaces) and `lightUrl` (light surfaces). Pass `query` or `set` to narrow it and get the keywords too. |
| Charts / data viz | `get_chart_palette`: categorical, sequential, semantic. Never a library-default rainbow. |
| Several token values at once | `list_tokens { refs: [...] }`: one call, not one per value. |
| The whole design system | `get_design_spec` without arguments is the section INDEX; fetch the slice you need with `get_section` or `part`. The full document is ~79 KB and rarely what you want. |
| Anything undefined | `get_fallback`: the derivation protocol. |
| Finding a rule | `search_design` (full-text) or `get_section`. |

## No MCP available?

Use the read URLs: https://galton-ai-assets.vercel.app/llms.txt (index) or https://galton-ai-assets.vercel.app/llms-full.txt (the whole brand in one file, good as uploaded Knowledge). Everything is open-CORS plain text/JSON.

No URL access either? Download the offline package: https://galton-ai-assets.vercel.app/bundle.zip. It unzips to a folder carrying this skill, every knowledge file, and every logo, icon and font as a real file. Its START-HERE.md is the map. Inside that folder, the absolute URL `https://galton-ai-assets.vercel.app/brand/X` and the local path `./brand/X` are the same file.