# Claude Design Brand > Detect the project's character from its repo (project type, audience, voice) and produce a complete brand foundation — palette, typography, mood, motion, voice, accessibility — with type-specific recommendations the user can accept or override. Writes docs/design/BRAND.md + .claude/design-system.md + a CLAUDE.md block. Run this FIRST. Trigger on /claude-design-brand or "set up brand", "init brand… - **Type:** Skill - **Install:** `agentstack add skill-erenisci-claude-design-prompter-claude-design-brand` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [erenisci](https://agentstack.voostack.com/s/erenisci) - **Installs:** 0 - **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [erenisci](https://github.com/erenisci) - **Source:** https://github.com/erenisci/claude-design-prompter/tree/master/claude-design-brand ## Install ```sh agentstack add skill-erenisci-claude-design-prompter-claude-design-brand ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Claude Design Brand The **first step** of the design pipeline. Reads the project, infers what kind of thing it is, and produces a brand foundation with opinionated recommendations. Output is the input to `claude-design-prompter` — every later artifact (logo, pages, landing) inherits these decisions. --- ## Localization User-facing strings (questions, options, summaries) MUST be in the language the user is conversing with you in **right now**. Detection: current conversation language → README/CLAUDE.md dominant language → English. Translate canonical English strings on the fly. Internal artifacts (file names, frontmatter keys, template keys) stay in English. --- ## Flow 1. **Deep scan (read-only)** to understand the project's character: - `README.md`, `CLAUDE.md`, `AGENTS.md`, `docs/`, `docs/architecture.md`, `docs/PRD.md`, `docs/adr/` - `package.json` (name, description, deps — auth lib, UI lib, ORM, etc.) - `requirements.txt` / `pyproject.toml` for backend hints - Existing palette/font hints in `tailwind.config.*`, `theme.*`, `tokens.json` - `.claude/design-system.md` if exists — those decisions are LOCKED 2. **Detect project type** using `project-types.md` (the static catalog of types + their signals + their recommendations). Common types: - hacker / cybersec / dev-tool, SaaS app, e-commerce, portfolio, marketing site, dashboard / data, docs, content/blog, fintech, social/community - Pick the single best match. If ambiguous, list the top two and let the user confirm. - Show what was detected and on what evidence: `Detected: — signals: `. The user can correct via the first question. 3. **Build a recommendation set from the type**: for each brand decision (mood, palette, typography, motion, voice, …), pull the type-specific recommended value from `project-types.md`. These appear as `(recommended)` in the AskUserQuestion options with a one-line WHY. 4. **Ask only what's missing** (LOCKED items skipped, ≤4 questions per `AskUserQuestion` call, recommended option always first and labeled). Round 1 is the most blocking 4; Round 2 is run only if essentials remain unresolved. Translate all strings to user language. Round 1 (max 4): - Project type (only if ambiguous; pre-checked recommendation visible) - Mood / character — recommendation comes from type - Default theme — recommendation from type (e.g. hacker → Dark; e-commerce → Light) - Primary palette direction — recommendation from type (e.g. hacker → terminal green mono; SaaS → indigo cool) Round 2 (only if still ambiguous): - Typography family direction — recommendation from type (mono for hacker, sans-serif geometric for SaaS, serif editorial for portfolio) - Motion intensity — recommendation from type - Voice / tone — recommendation from type - Special-effect call (e.g. matrix-rain for hacker, parallax for marketing) — recommendation; mark "off" by default unless type strongly implies it For anything unresolved beyond the asked rounds, apply the type's recommendation and mark `(auto)` in the output so the user can audit. 5. **Print a Claude Design prompt block** (fenced, copy-able) describing the brand. This is what the user pastes into Claude Design to generate brand visuals (mood board, type sample, palette swatches). Structure: ```markdown # — Brand Foundation ## Project & character - Type: - Audience: - Voice: - Mood: ## Palette - Primary: - Accent: - Neutrals: - Semantic: success / warning / danger / info (desaturated to fit mood) ## Typography - Heading: - Body: - Mono: - Scale: 12 / 14 / 16 / 18 / 20 / 24 / 30 / 36 ## Shape & motion - Radius character: - Elevation: - Motion intensity: ; durations 150 / 250 ms; ease-out entrance ## Special character ## Accessibility WCAG AA, reduced-motion respected. ``` 6. **Persist**: - `docs/design/BRAND.md` from `templates/BRAND.md.tmpl` (human-readable, prose). - `.claude/design-system.md` — create or update; this skill owns the `## Brand` section. Other sections (Logo, Pages, Landing) are owned by the prompter skill. - `CLAUDE.md` — inject (or update, marker-delimited, never duplicate) a "Design system" block pointing at `docs/design/` and `.claude/design-system.md`. If `CLAUDE.md` doesn't exist, create it with a header + the design block. - `docs/design/README.md` — index file; create or update entries. 7. **End with**: - Brand prompt block (above) for the user to paste into Claude Design. - One-line summary: "Brand done. Next: `/claude-design-prompter` to generate Logo / All-Pages / Landing — all will inherit these brand decisions." - If any `(auto)` defaults remain in BRAND.md (Quick mode was used or rounds were skipped), call them out explicitly: `> N decisions were filled from project-type defaults and marked (auto). Run /claude-design-prompter → Answer to confirm or override individual ones, or re-run /claude-design-brand to revisit.` ## Persistence rules - `BRAND.md` is rendered from `templates/BRAND.md.tmpl`. Every value gets a real string; if the user didn't confirm and no recommendation applies, write `TBD` (never fabricate). - `(auto)` suffix marks values defaulted from `project-types.md` without explicit confirmation. The downstream prompter modes propagate this flag (see prompter SKILL.md `(auto) propagation`). - `.claude/design-system.md` — own ONLY the `## Brand` section. Do not touch `## Logo`, `## Pages.*`, `## Landing` (those are owned by prompter). - `docs/design/README.md` — render from `templates/docs-readme.md.tmpl`. List every file currently in `docs/design/` (initially just `BRAND.md`). - `CLAUDE.md` — render `templates/claude-md-section.md.tmpl` between the markers; create the file if missing. Idempotent. --- ## Safety - Never overwrite `BRAND.md` without confirmation (offer merge / overwrite / skip). - Never invent brand assets the user hasn't approved. Unknown values → `TBD`, never fabricate hex codes or font choices outside the recommendation set. - Recommendations are opinions from `project-types.md` — always show WHY, never present as absolutes. - Translate user-facing strings; keep internal keys English. --- ## Agent Usage On invocation, the FIRST action is the deep scan + type detection (no AskUserQuestion yet — gather evidence first). Then present the detection in 2–4 lines and immediately move to Round 1 questions. No preamble like "let me scan the repo" — just do it and report findings together with the first question. If `.claude/design-system.md` already has a `## Brand` section: ask once whether to "View existing / Update specific decisions / Re-run from scratch". Default to View. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [erenisci](https://github.com/erenisci) - **Source:** [erenisci/claude-design-prompter](https://github.com/erenisci/claude-design-prompter) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-erenisci-claude-design-prompter-claude-design-brand - Seller: https://agentstack.voostack.com/s/erenisci - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.