viepilot 2.23.0 → 2.45.2

package/docs/brainstorm/session-2026-04-25.md

@@ -0,0 +1,109 @@
+# Brainstorm Session — 2026-04-25
+
+## Meta
+- **Session ID**: session-2026-04-25
+- **Topic**: Design.MD × ViePilot Integration (ENH-076)
+- **Status**: Completed
+- **workflow_version**: 2.43.2
+- **Participants**: User + AI
+
+## Summary
+Research session trên tiêu chuẩn Design.MD (Google Labs, April 2026) và brainstorm
+cách tích hợp nó toàn diện vào ViePilot workflow — từ brainstorm → crystallize → auto.
+
+## Research: Design.MD Standard
+
+Design.MD là file specification từ Google Labs (21/04/2026, Apache 2.0) — Markdown + YAML
+front matter mô tả visual design system để AI coding agents hiểu brand identity.
+
+- **Creator**: Google Labs / Google Stitch 2.0
+- **Format**: YAML front matter (tokens) + Markdown sections (rationale)
+- **Adopted by**: Claude Code, Cursor, Windsurf, Kiro
+- **Community**: awesome-design-md — 55+ brand examples
+
+### Standard sections
+- `colors`: semantic palette (primary, surface, accent, error, success, warning)
+- `typography`: fontFamily, fontSize, fontWeight, lineHeight
+- `spacing`: base unit + scale
+- `rounded`: corner radius tokens
+- `components`: component-specific tokens
+
+---
+
+## Decisions
+
+### D1: Scope — 5 integration points across ViePilot
+1. **vp-brainstorm**: auto-extract design tokens → generate design.md in session dir
+2. **vp-crystallize**: mandatory-acknowledge gate + export to project root
+3. **vp-auto**: preflight reads design.md, injects TOKEN_MAP into UI tasks
+4. **new skill vp-design**: init/sync/audit/import commands
+5. **Architect workspace**: design.html visual page
+
+### D2: vp-brainstorm token extraction
+- Triggered when `--ui` flag OR ≥2 design keywords detected
+- tokens extracted to `notes.md ## design_tokens` YAML section
+- design.md written to `.viepilot/ui-direction/{session-id}/design.md`
+
+### D3: vp-crystallize gate — mandatory-acknowledge (not hard block)
+- Trigger: design.md exists OR design_tokens in notes.md OR ≥3 design keywords
+- AUQ with 3 options: Export (recommended) / Finalize & export / Skip with note
+- Skip writes `design_md_status: skipped` (idempotent — won't re-ask same session)
+- Export path: Step 1D.14 — after existing coverage exports
+- Output: project root design.md + ARCHITECTURE.md ## Design System + PROJECT-CONTEXT.md flag
+
+### D4: vp-auto — 3 injection levels
+- Level 1: Silent context injection (always for UI tasks)
+- Level 2: Checklist items (when task has explicit UI acceptance criteria)
+- Level 3: Post-task compliance audit (when task output is HTML file)
+- UI keyword detection: html/css/style/component/tailwind/layout/button/card/form/page etc.
+- Edge cases: backend-only tasks skip, monorepo nearest-design.md wins
+
+### D5: vp-design skill commands
+- `--init`: Q&A from scratch OR import from awesome-design-md (55+ brands)
+- `--sync`: tokens → Tailwind config / CSS variables / SCSS
+- `--audit`: scan implementation vs spec, report deviations
+- `--import [brand]`: fetch community template (Linear, Notion, Stripe...)
+
+### D6: ARCHITECTURE.md Design System section format
+- Token summary table (colors/typography/spacing/rounded)
+- Design principles from design.md ## Overview
+- Reference to design.md at project root
+
+---
+
+## Open Questions
+
+- Q1: Should design.md be committed to git by default? (Yes — it's a project artifact like ARCHITECTURE.md)
+- Q2: For vp-design --sync with Tailwind: CSS custom properties or direct tailwind.config.js extension?
+  → Probably both, detected by project structure
+- Q3: awesome-design-md catalog integration — fetch at runtime vs. bundle with ViePilot?
+  → Fetch at runtime (keep ViePilot lean)
+
+---
+
+## Phases
+
+### Phase 1 (ENH-076 core)
+- 76.1: vp-brainstorm design token extraction + design.md generation
+- 76.2: vp-crystallize Step 1D.14 mandatory gate + export pipeline
+- 76.3: vp-auto preflight 5.5 + TOKEN_MAP injection + post-task audit
+- 76.4: new skill vp-design (SKILL.md + workflows/design.md)
+- 76.5: Architect workspace design.html page
+
+### Phase 2 (polish)
+- 76.6: awesome-design-md runtime catalog integration
+- 76.7: vp-design --sync multi-target (Tailwind + CSS vars + SCSS)
+- 76.8: vp-design --audit full HTML/CSS scan
+
+---
+
+## Action Items
+- [ ] Log ENH-076 via /vp-request
+- [ ] Plan phases via /vp-evolve
+- [ ] Execute via /vp-auto
+
+---
+
+## Project Meta Intake (FEAT-009)
+- Profile already bound — skip intake.
+

package/docs/skills-reference.md

@@ -525,3 +525,25 @@ Place custom `.pptx`/`.docx` files in `.viepilot/proposal-templates/` to overrid
 ```
 
 Full guide: `docs/user/features/proposal.md`
+
+---
+
+## /vp-skills
+
+**Purpose**: Agent-native global skill registry management — scan, list, install, uninstall, update, and inspect third-party skills from any project directory (ENH-062).
+
+### Commands
+| Command | Description |
+|---------|-------------|
+| `/vp-skills scan` | Refresh `~/.viepilot/skill-registry.json` by scanning all adapter dirs |
+| `/vp-skills list` | Display indexed skills table |
+| `/vp-skills install <src>` | Install from npm / `github:<user>/<repo>` / local path |
+| `/vp-skills uninstall <id>` | Remove skill from all adapter directories |
+| `/vp-skills update <id>` | Re-install from original source (`skill-meta.json`) |
+| `/vp-skills info <id>` | Show capabilities, best_practices, and adapter paths |
+
+### Notes
+- Uses installed vp-tools binary (`~/.claude/viepilot/bin/vp-tools.cjs`) — works from any project directory.
+- Fallback: `~/.cursor/viepilot/bin/vp-tools.cjs` for Cursor adapter.
+- Registry file: `~/.viepilot/skill-registry.json` — shared across all projects.
+- Confirmation AUQ prompt before destructive `uninstall` (Claude Code terminal only).

package/docs/user/features/adapters.md

@@ -8,7 +8,7 @@ ViePilot supports multiple AI coding platforms via its adapter system (FEAT-013)
 |------------|----------|------------|--------------|-------|--------------|
 | `claude-code` | Claude Code *(default)* | `~/.claude/skills/` | `~/.claude/viepilot/` | ✅ Stop, PreToolUse, … | `/vp-status` |
 | `cursor-agent` / `cursor-ide` | Cursor | `~/.cursor/skills/` | `~/.cursor/viepilot/` | — | `/vp-status` |
-| `antigravity` | Google Antigravity | `~/.antigravity/skills/` | `~/.antigravity/viepilot/` | — | `/vp-status` |
+| `antigravity` | Google Antigravity | `~/.gemini/antigravity/skills/` | `~/.gemini/antigravity/viepilot/` | — | `/vp-status` |
 | `codex` | OpenAI Codex CLI | `~/.codex/skills/` | `~/.codex/viepilot/` | — | `$vp-status` |
 | `copilot` | GitHub Copilot | `~/.config/gh-copilot/skills/` | `~/.config/gh-copilot/viepilot/` | — | `/vp-status` |
 
@@ -47,7 +47,7 @@ Skill source files use the neutral placeholder `{envToolDir}` in `execution_cont
 At install time, `{envToolDir}` is replaced with each adapter's `executionContextBase`:
 - `claude-code` → `.claude/viepilot`
 - `cursor` → `.cursor/viepilot`
-- `antigravity` → `.antigravity/viepilot`
+- `antigravity` → `.gemini/antigravity/viepilot`
 - `codex` → `.codex/viepilot`
 - `copilot` → `.config/gh-copilot/viepilot`
 

package/docs/user/features/scaffold-first.md

@@ -0,0 +1,62 @@
+# Scaffold-First Gate (BUG-020)
+
+`vp-auto` enforces a **scaffold-first** rule for framework-based projects: when a task initializes a new project, the canonical scaffold command MUST run before any framework-generated files are created manually.
+
+## How It Works
+
+1. **Detection** — vp-auto detects "project setup" tasks by keywords in the task title/objective: "set up", "initialize", "create project", "scaffold", "bootstrap", "new project"
+2. **Marker check** — checks for a framework marker file (e.g. `artisan` for Laravel, `manage.py` for Django) to determine if the project is already scaffolded
+3. **Scaffold command** — if not yet scaffolded, runs the scaffold command from either:
+   - `~/.viepilot/stacks/{stack}/SUMMARY.md` → `## Scaffold` → `init_command:` field (takes priority)
+   - The built-in heuristic table in `workflows/autonomous.md`
+4. **Block list** — refuses to create framework-native files (artisan, next.config.*, manage.py, nest-cli.json, pom.xml, etc.) without prior scaffold
+
+## Adding a Scaffold Command to a Stack
+
+In `~/.viepilot/stacks/{your-stack}/SUMMARY.md`, append:
+
+```markdown
+## Scaffold
+
+init_command: composer create-project laravel/laravel {name}
+marker_file: artisan
+```
+
+Replace `{name}` with the project directory name placeholder — vp-auto substitutes the actual project name at runtime.
+
+## Built-in Supported Stacks
+
+| Stack | Scaffold command | Marker file |
+|-------|-----------------|-------------|
+| laravel | `composer create-project laravel/laravel {name}` | `artisan` |
+| laravel-php84 | `composer create-project laravel/laravel:^11 {name}` | `artisan` |
+| nextjs | `npx create-next-app@latest {name}` | `next.config.js` |
+| nextjs-tailwind-shadcn-threejs | `npx create-next-app@latest {name} --typescript --tailwind --eslint --app` | `next.config.js` |
+| nestjs | `npx @nestjs/cli new {name}` | `nest-cli.json` |
+| rails | `rails new {name}` | `config/application.rb` |
+| django | `django-admin startproject {name} .` | `manage.py` |
+| spring-boot* | `spring init --dependencies=web,data-jpa,validation {name}` | `pom.xml` |
+| nuxt / vuejs* | `npx nuxi@latest init {name}` | `nuxt.config.ts` |
+| react | `npx create-react-app {name}` | `src/index.js` |
+| electron | `npx create-electron-app {name}` | `electron-builder.yml` |
+
+## Never-Handcraft Block List
+
+The following files must never be created manually in a project setup task without first running the scaffold command:
+
+- `artisan`, `composer.json` (Laravel)
+- `next.config.js`, `next.config.ts`, `pages/_app.*`, `app/layout.*` (Next.js)
+- `nest-cli.json`, `src/main.ts` + `tsconfig.json` combo (NestJS)
+- `manage.py`, `wsgi.py`, `asgi.py` (Django)
+- `config/application.rb`, `config/routes.rb`, `Gemfile` (Rails)
+- `pom.xml` or `build.gradle` with Spring Boot starters
+
+## Bypassing the Gate
+
+To override the gate (e.g. the project is already partially set up), add to the task file under `## Implementation Notes`:
+
+```
+scaffold_gate_waiver: project already scaffolded — artisan exists at {path}
+```
+
+Or at execution time, select "confirm project already scaffolded" at the control-point prompt.

package/docs/user/features/skill-registry.md

@@ -0,0 +1,125 @@
+# Skill Registry
+
+ViePilot can discover and index skills installed across all adapter directories. The registry powers workflow skill integration — brainstorm silently applies relevant skill best practices during UI-Direction sessions, crystallize locks skill decisions, and vp-auto executes with no re-prompting.
+
+## Commands
+
+| Command | Description |
+|---------|-------------|
+| `vp-tools scan-skills` | Scan all adapter dirs and write `~/.viepilot/skill-registry.json` |
+| `vp-tools list-skills` | Display indexed skills from the registry |
+| `vp-tools install-skill <source>` | Install a third-party skill *(Phase 91 — coming soon)* |
+
+## Extended SKILL.md Format
+
+Third-party skills (and built-in vp-* skills) can add optional metadata sections to their `SKILL.md`:
+
+```markdown
+## Capabilities
+- ui-generation
+- component-design
+- responsive-layout
+
+## Tags
+ui, design, frontend, css, html
+
+## Best Practices
+- Mobile-first: design breakpoints from 320px up
+- Use design tokens for consistency across components
+- BEM naming convention for CSS class names
+```
+
+### Section rules
+
+| Section | Format | Description |
+|---------|--------|-------------|
+| `## Capabilities` | One capability per line (`- item`) | Machine-readable ability tags used for skill matching |
+| `## Tags` | Comma-separated or one per line | Human-readable keywords for discovery |
+| `## Best Practices` | One practice per line (`- item`) | Applied silently when skill is matched in brainstorm/vp-auto |
+
+All sections are **optional** — skills without them are still indexed and available.
+
+### Backward compatibility
+
+Existing `SKILL.md` files that do not contain these sections continue to work without modification. The scanner indexes them with empty `capabilities`, `tags`, and `best_practices` arrays and marks them as `(legacy — no capabilities)` in `list-skills` output.
+
+## Scanner
+
+`vp-tools scan-skills` traverses the following adapter directories:
+
+| Adapter | Skill directory |
+|---------|----------------|
+| Claude Code | `~/.claude/skills/` |
+| Cursor | `~/.cursor/skills/` |
+| Codex | `~/.codex/skills/` |
+| Antigravity | `~/.gemini/antigravity/skills/` |
+| GitHub Copilot | `~/.config/gh-copilot/skills/` |
+
+Adapter directories that do not exist are silently skipped. The same skill found in multiple adapter directories is merged into a single registry entry with all adapters listed.
+
+## Registry Format
+
+The registry is written to `~/.viepilot/skill-registry.json`:
+
+```json
+{
+  "version": "1.0",
+  "last_scan": "2026-04-20T10:30:00.000Z",
+  "scan_paths": [
+    "/Users/you/.claude/skills",
+    "/Users/you/.cursor/skills"
+  ],
+  "skills": [
+    {
+      "id": "frontend-design",
+      "name": "frontend-design",
+      "source": null,
+      "version": null,
+      "description": "Opinionated frontend design guidance for UI-Direction sessions.",
+      "capabilities": ["ui-generation", "component-design", "responsive-layout"],
+      "tags": ["ui", "design", "frontend", "css"],
+      "best_practices": [
+        "Mobile-first: design breakpoints from 320px up",
+        "Use design tokens for consistency"
+      ],
+      "adapters": ["claude-code", "cursor"],
+      "installed_paths": {
+        "claude-code": "/Users/you/.claude/skills/frontend-design",
+        "cursor": "/Users/you/.cursor/skills/frontend-design"
+      }
+    }
+  ]
+}
+```
+
+### Schema reference
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `id` | string | Directory name (unique per skill) |
+| `name` | string | Display name (defaults to `id`) |
+| `source` | string \| null | Installation source: `npm:pkg`, `github:org/repo`, `./path`, or `null` for manually placed skills |
+| `version` | string \| null | Version from `skill-meta.json` (future); `null` for legacy skills |
+| `description` | string | First non-heading paragraph from `SKILL.md` |
+| `capabilities` | string[] | From `## Capabilities` section |
+| `tags` | string[] | From `## Tags` section |
+| `best_practices` | string[] | From `## Best Practices` section |
+| `adapters` | string[] | Adapter IDs where this skill is installed |
+| `installed_paths` | object | `{ adapterId: absolutePath }` |
+
+## Workflow Integration
+
+The registry is used automatically by ViePilot workflows:
+
+### Brainstorm (UI-Direction)
+When UI signals are detected during a brainstorm session, relevant skills (matched by `capabilities`) are silently applied — their `best_practices` are included in HTML generation without prompting. Matched skills are recorded in `notes.md` under `## skills_used`.
+
+### Crystallize (skill decision lock)
+After the brainstorm scope is locked, crystallize presents the skills used and asks the user to confirm which are required vs optional per phase. The decision is written to `PROJECT-CONTEXT.md` under `## Skills`.
+
+### vp-auto (silent execution)
+vp-auto reads `## Skills` from `PROJECT-CONTEXT.md` and injects skill best practices per task — no re-asking. Decisions made at crystallize time are final.
+
+## Refreshing the Registry
+
+Run `vp-tools scan-skills` any time you install a new skill manually or after `vp-tools install-skill` (which runs it automatically). The registry is not auto-refreshed on session start.

package/lib/adapters/antigravity.cjs

@@ -6,10 +6,10 @@ const fs = require('fs');
 module.exports = {
   id: 'antigravity',
   name: 'Antigravity',
-  skillsDir:   (home) => path.join(home, '.antigravity', 'skills'),
-  viepilotDir: (home) => path.join(home, '.antigravity', 'viepilot'),
+  skillsDir:   (home) => path.join(home, '.gemini', 'antigravity', 'skills'),
+  viepilotDir: (home) => path.join(home, '.gemini', 'antigravity', 'viepilot'),
   // {envToolDir} in SKILL.md files resolves to this value at install time (ENH-035)
-  executionContextBase: '.antigravity/viepilot',
+  executionContextBase: '.gemini/antigravity/viepilot',
   // Post-install hint shown in "Next actions" after viepilot install
   postInstallHint: 'Open project and run /vp-status',
   hooks: {
@@ -29,6 +29,7 @@ module.exports = {
   ],
   isAvailable: (home) => {
     const h = home || os.homedir();
-    return fs.existsSync(path.join(h, '.antigravity'));
+    return fs.existsSync(path.join(h, '.gemini', 'antigravity'))
+      || fs.existsSync(path.join(h, '.antigravity'));
   }
 };

package/lib/domain-packs/ai-product.json

@@ -0,0 +1,33 @@
+{
+  "id": "ai-product",
+  "label": "AI Product / LLM Application",
+  "topic_priority": ["llm-integration", "rag-pipeline", "prompt-mgmt", "eval", "ux", "guardrails"],
+  "extra_topics": [
+    {
+      "id": "llm-integration",
+      "label": "LLM Integration",
+      "questions": ["LLM provider (OpenAI/Anthropic/Gemini/local)?", "Model selection strategy (cost vs quality)?", "Streaming vs batch response?", "Fallback chain if primary model fails?"]
+    },
+    {
+      "id": "rag-pipeline",
+      "label": "RAG Pipeline",
+      "questions": ["Document ingestion format (PDF/HTML/Markdown)?", "Chunking strategy (fixed/semantic/hierarchical)?", "Vector DB (Pinecone/Qdrant/pgvector/Chroma)?", "Embedding model?", "Retrieval method (semantic/hybrid/BM25)?"]
+    },
+    {
+      "id": "prompt-mgmt",
+      "label": "Prompt Management",
+      "questions": ["Prompt versioning strategy?", "Prompt storage (DB/Git/Langfuse)?", "System prompt vs user prompt separation?", "Prompt injection protection?"]
+    },
+    {
+      "id": "eval-feedback",
+      "label": "Eval & Feedback Loop",
+      "questions": ["Eval framework (RAGAS/DeepEval/custom)?", "Human feedback collection (thumbs up/down/rubric)?", "Evals in CI/CD pipeline?", "Latency + cost tracking per request?"]
+    }
+  ],
+  "phase_template": {
+    "name": "ai-product",
+    "phases": ["LLM Wrapper & Auth", "Prompt Engineering & RAG", "Eval Pipeline", "UX & Feedback Loop", "Guardrails & Cost Control"]
+  },
+  "architect_pages": ["llm-arch.html", "rag-flow.html"],
+  "stacks_hint": ["langchain", "openai", "anthropic", "pgvector", "fastapi", "nextjs", "redis"]
+}

package/lib/domain-packs/data-science.json

@@ -0,0 +1,33 @@
+{
+  "id": "data-science",
+  "label": "Data Science / ML",
+  "topic_priority": ["dataset", "eda", "model-training", "evaluation", "serving", "monitoring"],
+  "extra_topics": [
+    {
+      "id": "dataset",
+      "label": "Dataset Management",
+      "questions": ["Data sources (structured/unstructured/streaming)?", "Data versioning strategy (DVC/Delta Lake/LakeFS)?", "PII/sensitive data handling?", "Train/val/test split strategy?"]
+    },
+    {
+      "id": "model-versioning",
+      "label": "Model Versioning",
+      "questions": ["Model registry (MLflow/W&B/SageMaker)?", "Experiment tracking metadata?", "Artifact storage (S3/GCS/local)?"]
+    },
+    {
+      "id": "drift-monitoring",
+      "label": "Drift & Quality Monitoring",
+      "questions": ["Data drift detection (Evidently/WhyLogs/custom)?", "Model performance degradation alerts?", "Retraining trigger strategy?"]
+    },
+    {
+      "id": "eval-pipeline",
+      "label": "Evaluation Pipeline",
+      "questions": ["Offline eval metrics (AUC/F1/BLEU/ROUGE)?", "Human eval process?", "A/B testing framework for model versions?"]
+    }
+  ],
+  "phase_template": {
+    "name": "ml-pipeline",
+    "phases": ["Data Ingestion & EDA", "Feature Engineering & Training", "Evaluation & Registry", "Serving & API", "Monitoring & Retraining"]
+  },
+  "architect_pages": ["ml-pipeline.html", "data-lineage.html"],
+  "stacks_hint": ["python", "pandas", "pytorch", "fastapi", "mlflow", "postgresql", "redis"]
+}

package/lib/domain-packs/devops.json

@@ -0,0 +1,33 @@
+{
+  "id": "devops",
+  "label": "DevOps / Platform Engineering",
+  "topic_priority": ["infra", "ci-cd", "observability", "slo", "incident-mgmt", "security"],
+  "extra_topics": [
+    {
+      "id": "iac",
+      "label": "Infrastructure as Code",
+      "questions": ["IaC tool (Terraform/Pulumi/CloudFormation/CDK)?", "Cloud provider (AWS/GCP/Azure/multi)?", "State backend (S3/GCS/Terraform Cloud)?", "Module/stack organization strategy?"]
+    },
+    {
+      "id": "ci-cd",
+      "label": "CI/CD Pipeline",
+      "questions": ["CI platform (GitHub Actions/GitLab CI/CircleCI/Jenkins)?", "Deploy strategy (blue-green/canary/rolling)?", "Secrets management (Vault/AWS SSM/GitHub Secrets)?", "PR-level preview environments?"]
+    },
+    {
+      "id": "slo-sla",
+      "label": "SLO / SLA",
+      "questions": ["Availability target (99.9%/99.99%)?", "Latency SLOs (p95/p99)?", "Error budget policy?", "SLO monitoring tool (Datadog/Prometheus/Grafana)?"]
+    },
+    {
+      "id": "incident-mgmt",
+      "label": "Incident Management",
+      "questions": ["On-call rotation tool (PagerDuty/OpsGenie/Rootly)?", "Runbook format and storage?", "Post-mortem process?", "Alert fatigue mitigation strategy?"]
+    }
+  ],
+  "phase_template": {
+    "name": "infra-ops",
+    "phases": ["Infrastructure Foundation", "CI/CD & Automation", "Observability & Alerting", "Runbooks & Incident Response"]
+  },
+  "architect_pages": ["infra-topology.html", "slo.html"],
+  "stacks_hint": ["terraform", "docker", "kubernetes", "prometheus", "grafana", "github-actions"]
+}

package/lib/domain-packs/mobile.json

@@ -0,0 +1,33 @@
+{
+  "id": "mobile",
+  "label": "Mobile App",
+  "topic_priority": ["auth", "core-features", "offline-sync", "push-notifications", "app-store", "deep-links"],
+  "extra_topics": [
+    {
+      "id": "push-notifications",
+      "label": "Push Notifications",
+      "questions": ["Push provider (FCM/APNs/OneSignal/Expo)?", "Notification categories and opt-in flow?", "Deep link handling from notification tap?", "Background/foreground handling?"]
+    },
+    {
+      "id": "offline-sync",
+      "label": "Offline Sync",
+      "questions": ["Offline-first or online-required?", "Conflict resolution strategy (last-write-wins/CRDT/manual)?", "Local storage (SQLite/Realm/AsyncStorage/Core Data)?", "Sync trigger (background/explicit/realtime)?"]
+    },
+    {
+      "id": "app-store",
+      "label": "App Store Submission",
+      "questions": ["Target platforms (iOS/Android/both)?", "App review checklist (privacy policy, ATT prompt, entitlements)?", "CI/CD for store deployment (Fastlane/EAS/Bitrise)?"]
+    },
+    {
+      "id": "deep-links",
+      "label": "Deep Links & Universal Links",
+      "questions": ["URL scheme vs universal links?", "Deferred deep linking (install → open specific screen)?", "Attribution provider (Branch/AppsFlyer/Adjust)?"]
+    }
+  ],
+  "phase_template": {
+    "name": "mobile-launch",
+    "phases": ["Auth & Navigation", "Core Features", "Offline & Sync", "Store Submission & Launch"]
+  },
+  "architect_pages": ["mobile-flow.html", "push.html"],
+  "stacks_hint": ["react-native", "expo", "flutter", "firebase", "postgresql"]
+}

package/lib/domain-packs/web-saas.json

@@ -0,0 +1,33 @@
+{
+  "id": "web-saas",
+  "label": "Web SaaS",
+  "topic_priority": ["auth", "user-data", "api", "billing", "admin", "onboarding", "content-mgmt"],
+  "extra_topics": [
+    {
+      "id": "billing",
+      "label": "Billing & Subscriptions",
+      "questions": ["Payment provider (Stripe/Paddle/LemonSqueezy)?", "Pricing model (flat/usage/per-seat)?", "Free trial or freemium tier?", "Proration on plan change?"]
+    },
+    {
+      "id": "multi-tenant",
+      "label": "Multi-tenancy",
+      "questions": ["Tenant isolation: row-level, schema-per-tenant, or separate DB?", "How are tenants identified (subdomain/slug/org)?", "Cross-tenant data visibility rules?"]
+    },
+    {
+      "id": "onboarding",
+      "label": "User Onboarding",
+      "questions": ["Onboarding flow steps (email verify → profile → invite team)?", "Guided setup wizard?", "Empty-state design for new accounts?"]
+    },
+    {
+      "id": "trial-freemium",
+      "label": "Trial / Freemium",
+      "questions": ["Trial duration?", "Feature gating strategy (usage caps vs feature flags)?", "Upgrade CTA placement?"]
+    }
+  ],
+  "phase_template": {
+    "name": "lean-startup",
+    "phases": ["Auth & Identity", "Core Features", "Monetization", "Scale & Ops"]
+  },
+  "architect_pages": ["billing.html", "tenant.html"],
+  "stacks_hint": ["nextjs", "nestjs", "postgresql", "stripe", "redis", "tailwind"]
+}