get-shit-pretty 0.6.0 → 0.6.2

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "gsp",
-  "version": "0.6.0",
+  "version": "0.6.2",
   "description": "Design engineering system for AI coding agents. Brand identity + design projects, from strategy to code.",
   "author": {
     "name": "jubscodes",

package/README.md

@@ -183,6 +183,7 @@ Create marketing campaign assets — landing page copy, social media content, la
 | `/gsp:brand-strategy` | Define archetype, positioning, personality, voice, messaging |
 | `/gsp:brand-identity` | Create visual identity — logo, color, type |
 | `/gsp:brand-patterns` | Build design system — tokens, components |
+| `/gsp:brand-refine` | Surgical token and palette adjustments mid-project |
 
 ### Project
 
@@ -200,6 +201,7 @@ Create marketing campaign assets — landing page copy, social media content, la
 
 | Command | What it does |
 |---------|--------------|
+| `/gsp:accessibility-audit` | Full WCAG 2.2 AA accessibility audit |
 | `/gsp:add-reference` | Add reference material to a project |
 | `/gsp:doctor` | Check project health |
 | `/gsp:update` | Update GSP to latest version |
@@ -240,7 +242,7 @@ GSP works across all major AI coding tools. The installer converts Claude Code's
 
 | Feature | Claude Code | OpenCode | Gemini CLI | Codex CLI |
 |---------|:-----------:|:--------:|:----------:|:---------:|
-| Skills | 24 | 24 | 24 | 24 |
+| Skills | 30 | 30 | 30 | 30 |
 | Agents | 15 | 15 | 15 (experimental) | — |
 | Slash syntax | `/gsp:command` | `/gsp-command` | `/gsp:command` | `$gsp-command` |
 | Prompts + templates | Yes | Yes | Yes | Yes |
@@ -329,8 +331,7 @@ get-shit-pretty/
 ├── scripts/               Hook scripts and utilities
 ├── gsp/                   Source of truth for all content
 │   ├── agents/            15 subagents (gsp-*.md)
-│   ├── commands/gsp/      20 slash commands (backward compat)
-│   ├── skills/            24 skills (*/SKILL.md — primary)
+│   ├── skills/            30 skills (*/SKILL.md)
 │   ├── hooks/             Plugin-level hooks (hooks.json)
 │   ├── prompts/           12 agent system prompts
 │   ├── templates/         Config, state, brief, roadmap templates
@@ -343,7 +344,7 @@ get-shit-pretty/
 └── CLAUDE.md              AI agent instructions for this repo
 ```
 
-Skills take precedence over commands when both exist. The installer reads from `gsp/` and writes to each runtime's config directory.
+The installer reads from `gsp/` and writes to each runtime's config directory.
 
 ---
 

package/bin/install.js

@@ -605,6 +605,8 @@ function convertClaudeSkillToOpencode(content, skillName) {
     if (trimmed.startsWith('color:')) { inList = false; continue; }
     if (trimmed.startsWith('context:')) { inList = false; continue; }
     if (trimmed.startsWith('agent:')) { inList = false; continue; }
+    if (trimmed.startsWith('model:')) { inList = false; continue; }
+    if (trimmed.startsWith('effort:')) { inList = false; continue; }
     // Skip list items under dropped fields
     if (inList) {
       if (trimmed.startsWith('- ')) continue;
@@ -783,6 +785,8 @@ function convertClaudeSkillToGemini(content, skillName) {
     if (trimmed.startsWith('color:')) { inList = false; continue; }
     if (trimmed.startsWith('context:')) { inList = false; continue; }
     if (trimmed.startsWith('agent:')) { inList = false; continue; }
+    if (trimmed.startsWith('model:')) { inList = false; continue; }
+    if (trimmed.startsWith('effort:')) { inList = false; continue; }
     if (inList) {
       if (trimmed.startsWith('- ')) continue;
       else if (trimmed && !trimmed.startsWith('-')) inList = false;
@@ -858,6 +862,8 @@ function convertClaudeSkillToCodex(content, skillName) {
     if (trimmed.startsWith('color:')) { inList = false; continue; }
     if (trimmed.startsWith('context:')) { inList = false; continue; }
     if (trimmed.startsWith('agent:')) { inList = false; continue; }
+    if (trimmed.startsWith('model:')) { inList = false; continue; }
+    if (trimmed.startsWith('effort:')) { inList = false; continue; }
     if (inList) {
       if (trimmed.startsWith('- ')) continue;
       else if (trimmed && !trimmed.startsWith('-')) inList = false;
@@ -915,6 +921,7 @@ function copyOpencodeSkills(srcDir, destDir, pathPrefix) {
     content = content.replace(/\.\/\.claude\//g, './.opencode/');
     content = convertClaudeSkillToOpencode(content, skillName);
     fs.writeFileSync(path.join(skillDest, 'SKILL.md'), content);
+    copySiblingFiles(path.join(srcDir, dir.name), skillDest, pathPrefix);
     count++;
   }
 
@@ -946,6 +953,7 @@ function copyCodexSkillsFromSource(srcDir, destDir, pathPrefix) {
     content = content.replace(/~\/\.claude\//g, pathPrefix);
     content = convertClaudeSkillToCodex(content, skillName);
     fs.writeFileSync(path.join(skillDest, 'SKILL.md'), content);
+    copySiblingFiles(path.join(srcDir, dir.name), skillDest, pathPrefix);
     count++;
   }
   return count;
@@ -986,6 +994,7 @@ function copyGeminiSkills(srcDir, destDir, pathPrefix) {
     content = content.replace(/\.\/\.claude\//g, './.gemini/');
     content = convertClaudeSkillToGemini(content, skillName);
     fs.writeFileSync(path.join(skillDest, 'SKILL.md'), content);
+    copySiblingFiles(path.join(srcDir, dir.name), skillDest, pathPrefix);
     count++;
   }
   return count;
@@ -1030,6 +1039,24 @@ function copyAgents(srcDir, destDir, pathPrefix, runtime, { clean = false } = {}
  * Copy Claude Code skills (global install path — no body conversion, only path replacement).
  * Returns skill count.
  */
+/**
+ * Recursively copy sibling files in a skill directory (everything except SKILL.md).
+ * All sibling files are copied verbatim — no path replacement applied.
+ */
+function copySiblingFiles(srcDir, destDir, pathPrefix) {
+  for (const entry of fs.readdirSync(srcDir, { withFileTypes: true })) {
+    if (entry.name === 'SKILL.md') continue; // already handled by caller
+    const srcPath = path.join(srcDir, entry.name);
+    const destPath = path.join(destDir, entry.name);
+    if (entry.isDirectory()) {
+      fs.mkdirSync(destPath, { recursive: true });
+      copySiblingFiles(srcPath, destPath, pathPrefix);
+    } else {
+      fs.copyFileSync(srcPath, destPath);
+    }
+  }
+}
+
 function copyClaudeSkills(srcDir, destDir, pathPrefix) {
   fs.mkdirSync(destDir, { recursive: true });
 
@@ -1045,11 +1072,13 @@ function copyClaudeSkills(srcDir, destDir, pathPrefix) {
     if (!dir.isDirectory()) continue;
     const skillMd = path.join(srcDir, dir.name, 'SKILL.md');
     if (!fs.existsSync(skillMd)) continue;
-    const destSkillDir = path.join(destDir, dir.name);
+    const skillName = dir.name.startsWith('gsp-') ? dir.name : `gsp-${dir.name}`;
+    const destSkillDir = path.join(destDir, skillName);
     fs.mkdirSync(destSkillDir, { recursive: true });
     let content = fs.readFileSync(skillMd, 'utf8');
     content = content.replace(/~\/\.claude\//g, pathPrefix);
     fs.writeFileSync(path.join(destSkillDir, 'SKILL.md'), content);
+    copySiblingFiles(path.join(srcDir, dir.name), destSkillDir, pathPrefix);
     skillCount++;
   }
   return skillCount;

package/gsp/agents/gsp-identity-designer.md

@@ -1,8 +1,8 @@
 ---
 name: gsp-identity-designer
 description: Creates visual identity — logo, color, typography, imagery. Spawned by /gsp:brand-identity.
-tools: Read, Write, Edit, Bash, Grep, Glob, WebSearch, WebFetch
-maxTurns: 40
+tools: Read, Write, Edit, Bash, Grep, Glob, WebFetch
+maxTurns: 30
 permissionMode: acceptEdits
 color: magenta
 ---
@@ -30,12 +30,11 @@ You do NOT create strategy or voice. You receive those as input and translate th
 1. **Absorb inputs** — strategy chunks for strategic grounding, voice-and-tone for verbal-visual alignment, mood board for visual starting point
 1.5. **Seed from style base** — if style base preset files are provided, read the `.yml` for token values (palette, typography, spacing) and the `.md` for design philosophy. Use these as vocabulary, not constraint — adapt freely where the brand demands it. The preset is a starting palette and type direction, not a rulebook.
 2. **Design logo system** — 3 distinct directions, each expressing strategy differently. For each: concept, rationale (connects to archetype + positioning), variations, usage rules
-3. **Build color system** — primary, secondary, accent, semantic. Each color needs strategic rationale. Include Hex, RGB, Pantone, CMYK. Map dark mode. Calculate WCAG AA contrast.
-4. **Generate palettes** — use tints.dev API: `https://tints.dev/api/{colorName}/{hexWithout#}`. Store in `identity/palettes.json`
+3. **Build color system** — primary, secondary, accent, neutral, semantic. Each color needs strategic rationale. Include Hex and RGB. Map dark mode. Calculate WCAG AA contrast. Semantic colors (success, warning, error, info) are standard values — define them in color-system.md but do NOT generate tints.dev palettes for them.
+4. **Generate palettes** — use tints.dev API: `https://tints.dev/api/{colorName}/{hexWithout#}`. **Only for brand colors (primary, secondary, accent) and neutrals.** Store in `identity/palettes.json`
 5. **Define typography** — primary + secondary typefaces. Connect choices to voice: "We chose X because our voice is Y"
 6. **Specify imagery** — photography, illustration, iconography. Connected to archetype and brand essence
 7. **Show applications** — brand in context across key touchpoints
-8. **Outline brand book** — 20-page structure
 
 ## Quality Standards
 - Every visual decision traces to strategy: "We chose X because [archetype/positioning/voice]"
@@ -44,7 +43,7 @@ You do NOT create strategy or voice. You receive those as input and translate th
 </methodology>
 
 <output>
-Write 6 chunks + palettes.json + INDEX.md to the identity directory (path provided by the skill that spawned you).
+Write 5 chunks + palettes.json + INDEX.md to the identity directory (path provided by the skill that spawned you).
 
 Each chunk follows `references/chunk-format.md`.
 
@@ -53,7 +52,6 @@ Each chunk follows `references/chunk-format.md`.
 3. **`typography.md`** (~60-80 lines) — primary + secondary typefaces with rationale, type scale, responsive behavior
 4. **`imagery-style.md`** (~50-70 lines) — photography, illustration, iconography guidelines
 5. **`brand-applications.md`** (~50-70 lines) — key touchpoints showing brand in use
-6. **`brand-book.md`** (~40-50 lines) — 20-page outline
 
 ### palettes.json
 tints.dev OKLCH palettes in the identity directory.
@@ -71,7 +69,6 @@ tints.dev OKLCH palettes in the identity directory.
 | Typography | [typography.md](./typography.md) | ~{N} |
 | Imagery Style | [imagery-style.md](./imagery-style.md) | ~{N} |
 | Brand Applications | [brand-applications.md](./brand-applications.md) | ~{N} |
-| Brand Book | [brand-book.md](./brand-book.md) | ~{N} |
 | Palettes | [palettes.json](./palettes.json) | — |
 ```
 </output>

package/gsp/hooks/hooks.json

@@ -12,6 +12,60 @@
       }
     ],
     "SubagentStop": [
+      {
+        "matcher": "gsp-designer",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-designer agent just finished. Verify: (1) at least one design/screen-*.md chunk was written, (2) design/INDEX.md was written, (3) design/preview.html was written. Report any missing deliverables to the user."
+          }
+        ]
+      },
+      {
+        "matcher": "gsp-critic",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-critic agent just finished. Verify: (1) critique/critique.md was written with a heuristics score, (2) critique/prioritized-fixes.md was written, (3) critique/strengths.md was written. Report any missing deliverables to the user."
+          }
+        ]
+      },
+      {
+        "matcher": "gsp-identity-designer",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-identity-designer agent just finished. Verify: (1) identity/INDEX.md was written, (2) identity/color-system.md exists, (3) identity/typography.md exists, (4) identity/palettes.json exists. Report any missing deliverables to the user."
+          }
+        ]
+      },
+      {
+        "matcher": "gsp-pattern-architect",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-pattern-architect agent just finished. Verify: (1) tokens.json was written, (2) INDEX.md was written, (3) at least one foundations/ chunk exists. Report any missing deliverables to the user."
+          }
+        ]
+      },
+      {
+        "matcher": "gsp-scoper",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-scoper agent just finished. Verify: (1) brief/scope.md was written, (2) brief/target-adaptations.md was written, (3) brief/INDEX.md was written. Report any missing deliverables to the user."
+          }
+        ]
+      },
+      {
+        "matcher": "gsp-campaign-director",
+        "hooks": [
+          {
+            "type": "prompt",
+            "prompt": "The gsp-campaign-director agent just finished. Verify: (1) launch/campaign-strategy.md was written, (2) launch/INDEX.md was written. Report any missing deliverables to the user."
+          }
+        ]
+      },
       {
         "matcher": "gsp-builder",
         "hooks": [

package/gsp/references/questioning.md

@@ -5,10 +5,18 @@ Guide brief-gathering conversations — used by `/gsp:start` for both brand brie
 
 ---
 
+## Core Rules
+
+**One decision per question.** Every question must be its own `AskUserQuestion` call. Never group multiple questions into a single message. Ask one thing, wait for the answer, use it to inform the next question.
+
+**Never re-ask what the user already answered.** If information was gathered in a prior phase and is available in BRIEF.md or other artifacts, use it — don't ask again. If you need to validate stale information, confirm it ("I see X from earlier — still accurate?") rather than asking from scratch. Each phase reads the prior phase's output; the user should never feel like they're repeating themselves.
+
+---
+
 ## Question Flow
 
 ### Phase 1: Context (Who & Why)
-Start broad, then narrow.
+Start broad, then narrow. Each is a separate `AskUserQuestion`.
 
 1. **What** is this project? (app, website, rebrand, campaign)
 2. **Who** is it for? (audience demographics, psychographics)
@@ -41,11 +49,11 @@ Start broad, then narrow.
 
 ## Questioning Techniques
 
-### Progressive Disclosure
-Don't ask all 18 questions at once. Group into 3-4 conversational rounds:
-- Round 1: Context + Brand (questions 1-8)
-- Round 2: Scope + Constraints (questions 9-16)
-- Round 3: Success + Gaps (questions 17-18 + follow-ups)
+### One at a Time
+Ask each question as its own `AskUserQuestion` call. Wait for the answer before asking the next. Use each answer to decide whether to skip, reframe, or drill into the next question.
+
+### Adapt and Skip
+Don't follow the list rigidly. If an early answer reveals enough context, skip later questions. If an answer is surprising, follow up before moving on. The sequence is a guide, not a script.
 
 ### Follow-Up Patterns
 - **"Tell me more about..."** — When an answer is surface-level
@@ -66,9 +74,9 @@ Don't ask what you can infer:
 - Reserve prose questions for open-ended exploration where you genuinely don't know the option space
 
 ### When to Use `AskUserQuestion` vs Prose
-- **Use `AskUserQuestion`** for: picking between defined directions (archetypes, styles, brands), yes/no/which decisions, selecting from existing items (brands, projects), mood/tone/size preferences
-- **Use prose** for: open-ended creative input ("tell me about your brand"), gathering context you can't predict ("what problem does this solve?"), follow-up clarifications where the answer space is unbounded
-- **Rule of thumb:** if you can write 2-4 meaningful options with descriptions, use `AskUserQuestion`. If you'd be guessing at what the options should be, use prose.
+- **Always use `AskUserQuestion`** — this is the default for every user-facing question
+- **Prose is only for** follow-up clarifications where the answer space is truly unbounded and you cannot write meaningful options
+- **Rule of thumb:** if you can write 2-4 meaningful options with descriptions, use `AskUserQuestion`. If you'd be guessing at what the options should be, still use `AskUserQuestion` with an open-ended framing.
 
 ### Knowing When You Have Enough
 A brief is complete when you can answer:

package/gsp/skills/get-shit-pretty/SKILL.md

@@ -2,6 +2,8 @@
 name: get-shit-pretty
 description: "Design engineering for AI coding tools. Full pipeline: brand research, strategy, identity, design system, UI design, critique, accessibility audit, build, and launch. Runs specialized agents with Apple HIG, Nielsen's heuristics, WCAG 2.2 AA, and design token standards baked in."
 user-invocable: false
+model: opus
+effort: high
 ---
 
 # Get Shit Pretty

package/gsp/skills/gsp-accessibility/SKILL.md

@@ -2,6 +2,7 @@
 name: accessibility
 description: Quick contrast checks and token WCAG audits — inline, no agent
 user-invocable: true
+model: sonnet
 allowed-tools:
   - Read
   - Write
@@ -30,6 +31,7 @@ Run lightweight accessibility checks inline — contrast ratio lookups and token
 
 <rules>
 - Always use `AskUserQuestion` for user interaction — never prompt via plain text
+- One decision per question — never batch multiple questions in a single message
 - Quick check mode (`--check`) produces display output only — no files written
 - Token audit mode runs inline — no agent spawned
 - Default conformance level is AA unless overridden by `--level AAA` or config

package/gsp/skills/gsp-accessibility-audit/SKILL.md

@@ -2,6 +2,8 @@
 name: accessibility-audit
 description: Full WCAG accessibility audit — design screens, codebase, or generate compliance statement
 user-invocable: true
+model: opus
+effort: high
 allowed-tools:
   - Read
   - Write
@@ -37,6 +39,7 @@ Run full accessibility audits — design screen reviews, codebase ARIA/keyboard/
 
 <rules>
 - Always use `AskUserQuestion` for user interaction — never prompt via plain text
+- One decision per question — never batch multiple questions in a single message
 - Statement mode reads prior audit results — fails gracefully if none exist
 - Default conformance level is AA unless overridden by `--level AAA` or config
 - Foundation chunks follow `references/chunk-format.md` format

package/gsp/skills/gsp-add-reference/SKILL.md

@@ -2,6 +2,7 @@
 name: add-reference
 description: Add reference material to a project
 user-invocable: true
+model: sonnet
 allowed-tools:
   - Read
   - Write

package/gsp/skills/gsp-art/SKILL.md

@@ -2,6 +2,7 @@
 name: art
 description: "Craft ASCII art interactively — you direct, the artist creates"
 user-invocable: true
+model: sonnet
 allowed-tools:
   - Read
   - Bash

package/gsp/skills/gsp-brand-audit/SKILL.md

@@ -2,6 +2,8 @@
 name: brand-audit
 description: Audit an existing brand before evolving it
 user-invocable: true
+model: opus
+effort: high
 allowed-tools:
   - Read
   - Write
@@ -29,6 +31,7 @@ Audit an existing brand. Produce evolution map that guides research, strategy, a
 
 <rules>
 - Always use `AskUserQuestion` for user-facing questions — never raw text prompts
+- One decision per question — never batch multiple questions in a single message
 - Keep interactions concise — gather assets, confirm scope, spawn agent
 - Artifacts must balance human readability with agent consumption for downstream phases
 </rules>
@@ -43,13 +46,15 @@ Read `{BRAND_PATH}/config.json` to confirm `brand_mode` is `evolve`.
 
 If missing, tell user to run `/gsp:start` first.
 
-## Step 2: Gather brand assets
+## Step 2: Load brand assets from brief
 
-Use `AskUserQuestion` to prompt for assets:
+Read `{BRAND_PATH}/BRIEF.md` — brand assets were already gathered during `/gsp:start`. Extract any logos, colors, guidelines, URLs, or descriptions the user provided.
+
+If the brief has no asset information (legacy or incomplete brief), use `AskUserQuestion`:
 - **Share assets now** — "I have guidelines, colors, fonts, voice samples, or URLs to share"
 - **Describe the brand** — "I'll describe it in my own words"
 
-Accept whatever format. Infer from partial info. If URLs provided, use WebFetch. Don't over-ask — work with what's given.
+If URLs were provided (in brief or just now), use WebFetch. Don't re-ask for information already in the brief.
 
 ## Step 3: Spawn auditor
 

package/gsp/skills/gsp-brand-identity/SKILL.md

@@ -2,6 +2,8 @@
 name: brand-identity
 description: Create your visual identity — logo, color, typography
 user-invocable: true
+model: opus
+effort: high
 allowed-tools:
   - Read
   - Write
@@ -31,6 +33,8 @@ Build the brand's visual identity.
 
 <rules>
 - Always use `AskUserQuestion` for user-facing questions — never raw text prompts
+- One decision per question — never batch multiple questions in a single message
+- Never re-ask what the user already answered in a prior phase — read BRIEF.md and strategy chunks, build on them
 - Every visual decision must trace to strategy — archetype, positioning, or voice
 - Artifacts must balance human readability with agent consumption for downstream phases
 </rules>
@@ -75,32 +79,34 @@ Use `AskUserQuestion` with 2-3 visual directions:
 - **Description:** color palette direction, typography feel, overall aesthetic
 - **Preview:** "Palette: {key colors}. Type: {typeface style}. Feel: {1-line vibe}."
 
-After selection, ask for hard constraints or visual references via `AskUserQuestion`:
+## Step 2b: Constraints
+
+After visual direction is confirmed, ask as a separate `AskUserQuestion`:
 - **No constraints** — "Go ahead with this direction"
-- **Add constraints** — "I have specific requirements"
+- **Add constraints** — "I have specific requirements (colors to avoid, accessibility needs, existing assets to match)"
 
 ## Step 3: Spawn identity designer
 
-Spawn the `gsp-identity-designer` agent with:
-- BRIEF.md content
-- Strategy chunks: archetype.md, positioning.md, brand-platform.md, voice-and-tone.md
-- discover/mood-board-direction.md
-- Brand Identity Creator prompt (02)
-- Identity output template
+Spawn the `gsp-identity-designer` agent. **Inline all content** — the agent should not need to read any input files.
+
+Pass in the agent prompt:
+- **Content of** BRIEF.md (loaded in Step 1)
+- **Content of** strategy chunks: archetype.md, positioning.md, brand-platform.md, voice-and-tone.md (loaded in Step 1)
+- **Content of** discover/mood-board-direction.md (loaded in Step 1)
+- **Content of** style base preset files `.yml` + `.md` (when loaded in Step 1)
+- **Content of** audit/brand-inventory.md (when loaded in Step 2)
+- Brand Identity Creator prompt (02), identity output template, color composition reference (from execution_context)
 - User-confirmed visual direction + constraints
-- Style base preset files (`.yml` + `.md`) if loaded
-- Audit chunks if they exist
 - **Output path:** `{BRAND_PATH}/identity/`
 
-The agent writes 6 chunks + palettes.json + INDEX.md:
+The agent writes 5 chunks + palettes.json + INDEX.md:
 1. `logo-directions.md`
 2. `color-system.md`
 3. `typography.md`
 4. `imagery-style.md`
 5. `brand-applications.md`
-6. `brand-book.md`
-7. `palettes.json`
-8. `INDEX.md`
+6. `palettes.json`
+7. `INDEX.md`
 
 ## Step 4: Perspective check
 

package/gsp/skills/gsp-brand-patterns/SKILL.md

@@ -2,6 +2,8 @@
 name: brand-patterns
 description: Build your design system — tokens, components, brand guidelines
 user-invocable: true
+model: opus
+effort: high
 allowed-tools:
   - Read
   - Write
@@ -30,6 +32,11 @@ Build the design system, generate brand guidelines, and complete the branding di
 @${CLAUDE_SKILL_DIR}/../../references/design-tokens.md
 </execution_context>
 
+<rules>
+- Always use `AskUserQuestion` for user-facing questions — never raw text prompts
+- One decision per question — never batch multiple questions in a single message
+</rules>
+
 <process>
 ## Step 0: Resolve brand
 
@@ -66,14 +73,20 @@ If `style_base` is empty or missing, load `${CLAUDE_SKILL_DIR}/../gsp-style/styl
 
 **Always scan:** If `.design/system/` docs don't exist, invoke `/gsp:design-system` via Skill tool to scan the codebase. If they already exist, read them. Either way, load STACK.md, COMPONENTS.md, and TOKENS.md before continuing.
 
-Then ask the user:
-1. "Will this brand target a specific tech stack?" (React + Tailwind, React Native + NativeWind, vanilla CSS, etc.)
-   - If the scan detected a stack, present it as the default: "I see you're using {framework} with {styling} — build on that?"
-   - Store answer in `{BRAND_PATH}/config.json` → `system_config.tech_stack`
-2. Based on scan results, determine system strategy:
-   - If scan found existing tokens/components: "You have an existing design system. Want to evolve it, rethink it from scratch, or ignore it?"
-   - Store strategy in `{BRAND_PATH}/config.json` → `system_config.system_strategy`
-   - If scan found no tokens/components (greenfield/boilerplate): default to `generate`, skip the question
+Then ask the user (each as its own `AskUserQuestion`):
+
+1. Tech stack — if the scan detected a stack, use `AskUserQuestion`:
+   - **Yes, build on {framework} + {styling}** — "Use what's already here"
+   - **Different stack** — "I want to target a different tech stack"
+   If no stack detected, use open-ended `AskUserQuestion`: "What tech stack will this brand target?"
+   Store answer in `{BRAND_PATH}/config.json` → `system_config.tech_stack`
+
+2. System strategy — only ask if scan found existing tokens/components. Use `AskUserQuestion`:
+   - **Evolve** — "Extend the existing design system"
+   - **Rethink** — "Redesign from scratch, informed by what exists"
+   - **Ignore** — "Start fresh, don't reference the existing system"
+   Store strategy in `{BRAND_PATH}/config.json` → `system_config.system_strategy`
+   If scan found no tokens/components (greenfield/boilerplate): default to `generate`, skip this question
 
 ## Step 2: Determine system strategy
 
@@ -121,37 +134,37 @@ The agent writes foundations only:
 
 ## Step 3.5: Foundation review (interactive)
 
-Read the foundation outputs and present:
+Read the foundation outputs and present a compact summary:
 
-"Here are the design system foundations:
+"Design system foundations:
  Color: {semantic mapping summary from foundations/color-system.md}
  Typography: {type scale summary from foundations/typography.md}
  Spacing: 8px base → {scale from foundations/spacing.md}
- Design principles: {list from principles.md}
+ Design principles: {list from principles.md}"
 
- Everything right? Adjustments before building components?"
+Use `AskUserQuestion`:
+- **Looks good** — "Build components on these foundations"
+- **Adjust colors** — "I want to tweak the color system"
+- **Adjust typography** — "I want to tweak the type scale"
+- **Adjust other** — "I want to change spacing, principles, or other foundations"
 
-Wait for user input. If adjustments needed, update the relevant foundation chunks.
+If adjustments needed, update the relevant foundation chunks, then re-present.
 
 ## Step 3.75: Perspective check
 
 Before building components, load persona profiles from `{BRAND_PATH}/BRIEF.md` and present stakeholder reactions:
 
-"Before we build the component library on these foundations:
+"Before we build the component library:
 
- The Customer ({primary persona name from BRIEF.md}):
- "{would a user recognize this as {brand}? Does the system feel like the brand?}"
+ {primary persona name}: {would they recognize this as {brand}? Does it feel right?}
+ Skeptic: {are the tokens flexible enough? Are the principles actionable?}
+ {top competitor}: {how does this compare to industry standards?}"
 
- The Skeptic (internal stakeholder):
- "{challenges system decisions — are the tokens flexible enough? Are the principles actionable?}"
+Use `AskUserQuestion`:
+- **Build components** — "These foundations are solid, let's go"
+- **Adjust** — "One of these concerns resonates — I want to change something"
 
- The Competitor ({top competitor name}):
- "{how does this system compare to industry standards? Any gaps?}"
-
- Any of these concerns resonate? Want to adjust foundations before building components?"
-
-If user wants changes → update foundations.
-If confirmed → proceed to components pass.
+If adjust → update foundations and re-present. If confirmed → proceed to components pass.
 
 ## Step 4: Spawn pattern architect — Pass 2: Components
 
@@ -208,7 +221,9 @@ Tell the user: "Brand kit saved to `guidelines.html` — open it in your browser
 
 Render phase transition (see `references/phase-transitions.md`).
 
-**Critical:** When the user chooses "Start a project", invoke `/gsp:start` via the Skill tool. Do NOT attempt to handle project setup inline — `/gsp:start` has the codebase scanning, questioning rounds, and brief-writing logic needed for a proper project setup. The branding agent's context is spent on brand work and lacks the project setup methodology.
+**E2E mode:** Read `{BRAND_PATH}/config.json`. If `e2e` is `true`, auto-invoke `/gsp:start` via Skill tool — it will detect the completed brand and route directly to project setup (Step 4). No need to ask the user.
+
+**Non-E2E:** When the user chooses "Start a project", invoke `/gsp:start` via the Skill tool. Do NOT attempt to handle project setup inline — `/gsp:start` has the codebase scanning, questioning rounds, and brief-writing logic needed for a proper project setup. The branding agent's context is spent on brand work and lacks the project setup methodology.
 
 Also display a brand summary after the standard transition — this is the final branding phase:
 

package/gsp/skills/gsp-brand-refine/SKILL.md

@@ -2,6 +2,7 @@
 name: brand-refine
 description: Targeted brand adjustments mid-project — tweak colors, typography, or spacing without re-running the full branding diamond
 user-invocable: true
+model: sonnet
 allowed-tools:
   - Read
   - Write
@@ -31,6 +32,7 @@ Accept natural language feedback about brand visuals, identify which tokens are
 
 <rules>
 - Always use `AskUserQuestion` for user interaction — never prompt via plain text
+- One decision per question — never batch multiple questions in a single message
 - Never update artifacts without showing before/after and getting confirmation
 - Only touch tokens directly affected by the feedback
 - Preserve existing token structure — edit values in place, don't restructure

package/gsp/skills/gsp-brand-research/SKILL.md

@@ -2,6 +2,8 @@
 name: brand-research
 description: Research your market and competitors
 user-invocable: true
+model: sonnet
+effort: high
 allowed-tools:
   - Read
   - Write
@@ -31,6 +33,7 @@ Research market context that will inform brand strategy.
 
 <rules>
 - Always use `AskUserQuestion` for user-facing questions — never raw text prompts
+- One decision per question — never batch multiple questions in a single message
 - Keep interactions concise — 1-2 exchanges max before spawning the agent
 - Artifacts must balance human readability with agent consumption for downstream phases
 </rules>