diff --git a/.github/ISSUE_TEMPLATE/agent-task.yml b/.github/ISSUE_TEMPLATE/agent-task.yml
new file mode 100644
index 00000000..59004201
--- /dev/null
+++ b/.github/ISSUE_TEMPLATE/agent-task.yml
@@ -0,0 +1,117 @@
+name: Agent implementation task
+description: Define a bounded development task for an AI coding agent
+title: "[Agent Task]: "
+labels: []
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Read `AGENTS.md` and `docs/AGENT_BUILD_FLOW.md` before implementation. Do not include secrets or real identity documents.
+
+  - type: textarea
+    id: objective
+    attributes:
+      label: Objective
+      description: State one measurable user or business outcome.
+      placeholder: A public contact enquiry is stored before the visitor sees a success message.
+    validations:
+      required: true
+
+  - type: textarea
+    id: acceptance
+    attributes:
+      label: Acceptance criteria
+      description: List observable conditions that prove the task is complete.
+      placeholder: |
+        - Valid submissions are persisted.
+        - Invalid input returns 400.
+        - Storage failure returns 503.
+        - Tests cover success and failure paths.
+    validations:
+      required: true
+
+  - type: textarea
+    id: scope
+    attributes:
+      label: In-scope routes, roles, and files
+      description: Identify affected user roles, routes, models, or integrations.
+    validations:
+      required: true
+
+  - type: textarea
+    id: exclusions
+    attributes:
+      label: Explicit exclusions
+      description: State what the agent must not change or activate.
+      placeholder: |
+        - No paid service activation.
+        - No production credentials.
+        - No direct commits to main.
+        - No sensitive document upload.
+    validations:
+      required: true
+
+  - type: dropdown
+    id: data
+    attributes:
+      label: Sensitive-data level
+      options:
+        - None or public content
+        - Account or contact information
+        - Authentication or authorization
+        - Identity, financial, legal, or regulated information
+    validations:
+      required: true
+
+  - type: dropdown
+    id: billing
+    attributes:
+      label: Billing impact
+      options:
+        - None
+        - Uses an existing free or already-approved service
+        - May create metered usage
+        - Requires owner approval before implementation can finish
+    validations:
+      required: true
+
+  - type: textarea
+    id: dependencies
+    attributes:
+      label: Manual and provider dependencies
+      description: List accounts, approvals, provider decisions, or credentials the owner must supply.
+      placeholder: None, or describe the exact owner-only action.
+    validations:
+      required: true
+
+  - type: textarea
+    id: tests
+    attributes:
+      label: Required verification
+      description: Include unit, authorization, fail-closed, preview, and production checks as applicable.
+      placeholder: |
+        - pnpm run verify
+        - Vercel preview route check
+        - No external billable calls in tests
+    validations:
+      required: true
+
+  - type: textarea
+    id: rollback
+    attributes:
+      label: Rollback expectation
+      description: Explain how the change can be disabled or reverted safely.
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: guardrails
+    attributes:
+      label: Guardrails
+      options:
+        - label: I have not placed secrets, passwords, API keys, or real sensitive documents in this issue.
+          required: true
+        - label: The task does not authorize billing, legal claims, certifications, or regulated-service activation.
+          required: true
+        - label: The agent must use a branch and pull request.
+          required: true
diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md
new file mode 100644
index 00000000..5237f411
--- /dev/null
+++ b/.github/pull_request_template.md
@@ -0,0 +1,50 @@
+## Objective
+
+Closes #
+
+Describe the single measurable outcome delivered by this pull request.
+
+## Changes
+
+- 
+
+## Files and data models affected
+
+- 
+
+## Trust, security, privacy, and billing review
+
+- [ ] No secrets or real sensitive documents are included.
+- [ ] Public wording does not overstate live, verified, encrypted, certified, guaranteed, or continuous capability.
+- [ ] Authorization and validation boundaries were reviewed.
+- [ ] Missing provider configuration fails closed.
+- [ ] No paid plan, billing, or metered feature was activated.
+- [ ] Database changes include compatibility and rollback notes, or no schema change was made.
+
+## Verification
+
+- [ ] `pnpm run verify`
+- [ ] Relevant authorization and validation failures tested
+- [ ] Vercel preview checked
+- [ ] Build/runtime logs checked
+- [ ] No billable external calls made during testing
+
+Evidence:
+
+```text
+Paste concise command results or route/status checks here.
+```
+
+## Manual actions remaining
+
+State exactly what requires the business owner, provider account, legal approval, or production credential. Write `None` when no manual action remains.
+
+## Rollback
+
+Describe how to revert, disable, or fail closed safely.
+
+## Agent declaration
+
+- [ ] I read `AGENTS.md` and `docs/AGENT_BUILD_FLOW.md`.
+- [ ] This pull request stays within the linked issue scope.
+- [ ] I am not claiming completion while a required gate is failing.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index f2e3867f..1bde4dff 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,62 +1,45 @@
-name: CI (Tests & Validation Only - Vercel Handles Deployment)
+name: Manual Build Verification
 
 on:
-  pull_request:
-    branches: [main]
+  workflow_dispatch:
 
-jobs:
-  test:
-    name: Unit Tests
-    runs-on: ubuntu-22.04
+permissions:
+  contents: read
 
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
+concurrency:
+  group: manual-build-verification-${{ github.ref }}
+  cancel-in-progress: true
 
-      - name: Install pnpm
-        uses: pnpm/action-setup@v4
-        with:
-          version: 9
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v4
-        with:
-          node-version: 20
-          cache: pnpm
-
-      - name: Install dependencies
-        run: pnpm install --frozen-lockfile
-
-      - name: Generate Prisma client
-        run: pnpm run db:generate
-        env:
-          POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost:5432/gem_ci
-          POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost:5432/gem_ci
-
-      - name: Run unit tests
-        run: pnpm test
-        env:
-          NODE_ENV: test
-          JWT_SECRET: ci-test-secret-min-32-characters-long-placeholder
-          POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost:5432/gem_ci
-          POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost:5432/gem_ci
-
-  lint-and-build:
-    name: Lint & Build
+jobs:
+  verify:
+    name: Lint, Typecheck, Test, Build
     runs-on: ubuntu-22.04
-    needs: test
+    timeout-minutes: 30
+
+    env:
+      NODE_ENV: test
+      JWT_SECRET: ci-test-secret-min-32-characters-long-placeholder
+      POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost:5432/gem_ci
+      POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost:5432/gem_ci
+      NEXT_PUBLIC_APP_URL: https://www.gemcybersecurityassist.com
+      NEXT_PUBLIC_APP_NAME: GEM Enterprise
+      NEXT_PUBLIC_AI_DISCLOSURE_TEXT: GEM Concierge is an AI assistant. Responses are informational and require human review for sensitive decisions.
+      AUDIT_ENABLED: "true"
+      SMTP_HOST: ""
+      SMTP_PORT: "587"
+      SMTP_USER: ""
+      SMTP_PASS: ""
+      CRON_SECRET: ci-cron-secret
 
     steps:
       - name: Checkout
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
 
       - name: Install pnpm
-        uses: pnpm/action-setup@v4
-        with:
-          version: 9
+        uses: pnpm/action-setup@v6
 
       - name: Setup Node.js
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@v6
         with:
           node-version: 20
           cache: pnpm
@@ -64,28 +47,5 @@ jobs:
       - name: Install dependencies
         run: pnpm install --frozen-lockfile
 
-      - name: Generate Prisma client
-        run: pnpm run db:generate
-        env:
-          POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost:5432/gem_ci
-          POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost:5432/gem_ci
-
-      - name: Lint
-        run: pnpm run lint
-
-      - name: Build
-        run: pnpm run build
-        env:
-          POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost:5432/gem_ci
-          POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost:5432/gem_ci
-          JWT_SECRET: ${{ secrets.JWT_SECRET || 'ci-build-secret-min-32-characters-long' }}
-          NEXT_PUBLIC_APP_URL: https://gem-enterprise.vercel.app
-          NEXT_PUBLIC_APP_NAME: GEM Enterprise
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-          NEXT_PUBLIC_AI_DISCLOSURE_TEXT: ${{ secrets.NEXT_PUBLIC_AI_DISCLOSURE_TEXT || 'GEM Concierge is an AI assistant. Responses are for informational purposes only.' }}
-          AUDIT_ENABLED: 'true'
-          SMTP_HOST: ''
-          SMTP_PORT: '587'
-          SMTP_USER: ''
-          SMTP_PASS: ''
-          CRON_SECRET: ci-cron-secret
+      - name: Verify repository
+        run: pnpm run verify
diff --git a/.github/workflows/codeql.yml b/.github/workflows/codeql.yml
index d6cf71be..d5b6ae4e 100644
--- a/.github/workflows/codeql.yml
+++ b/.github/workflows/codeql.yml
@@ -1,89 +1,44 @@
-# CodeQL Analysis - Runs only on pull requests to avoid deployment interference
-name: "CodeQL Advanced"
+name: Manual CodeQL Analysis
 
 on:
-  pull_request:
-    branches: [ "main" ]
-  schedule:
-    - cron: '15 6 * * 4'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  actions: read
+  packages: read
+  security-events: write
+
+concurrency:
+  group: manual-codeql-${{ github.ref }}
+  cancel-in-progress: true
 
 jobs:
   analyze:
     name: Analyze (${{ matrix.language }})
-    # Runner size impacts CodeQL analysis time. To learn more, please see:
-    #   - https://gh.io/recommended-hardware-resources-for-running-codeql
-    #   - https://gh.io/supported-runners-and-hardware-resources
-    #   - https://gh.io/using-larger-runners (GitHub.com only)
-    # Consider using larger runners or machines with greater resources for possible analysis time improvements.
-    runs-on: ${{ (matrix.language == 'swift' && 'macos-latest') || 'ubuntu-latest' }}
-    permissions:
-      # required for all workflows
-      security-events: write
-
-      # required to fetch internal or private CodeQL packs
-      packages: read
-
-      # only required for workflows in private repositories
-      actions: read
-      contents: read
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
 
     strategy:
       fail-fast: false
       matrix:
         include:
-        - language: actions
-          build-mode: none
-        - language: javascript-typescript
-          build-mode: none
-        # CodeQL supports the following values keywords for 'language': 'actions', 'c-cpp', 'csharp', 'go', 'java-kotlin', 'javascript-typescript', 'python', 'ruby', 'rust', 'swift'
-        # Use `c-cpp` to analyze code written in C, C++ or both
-        # Use 'java-kotlin' to analyze code written in Java, Kotlin or both
-        # Use 'javascript-typescript' to analyze code written in JavaScript, TypeScript or both
-        # To learn more about changing the languages that are analyzed or customizing the build mode for your analysis,
-        # see https://docs.github.com/en/code-security/code-scanning/creating-an-advanced-setup-for-code-scanning/customizing-your-advanced-setup-for-code-scanning.
-        # If you are analyzing a compiled language, you can modify the 'build-mode' for that language to customize how
-        # your codebase is analyzed, see https://docs.github.com/en/code-security/code-scanning/creating-an-advanced-setup-for-code-scanning/codeql-code-scanning-for-compiled-languages
-    steps:
-    - name: Checkout repository
-      uses: actions/checkout@v4
+          - language: actions
+            build-mode: none
+          - language: javascript-typescript
+            build-mode: none
 
-    # Add any setup steps before running the `github/codeql-action/init` action.
-    # This includes steps like installing compilers or runtimes (`actions/setup-node`
-    # or others). This is typically only required for manual builds.
-    # - name: Setup runtime (example)
-    #   uses: actions/setup-example@v1
-
-    # Initializes the CodeQL tools for scanning.
-    - name: Initialize CodeQL
-      uses: github/codeql-action/init@v4
-      with:
-        languages: ${{ matrix.language }}
-        build-mode: ${{ matrix.build-mode }}
-        # If you wish to specify custom queries, you can do so here or in a config file.
-        # By default, queries listed here will override any specified in a config file.
-        # Prefix the list here with "+" to use these queries and those in the config file.
-
-        # For more details on CodeQL's query packs, refer to: https://docs.github.com/en/code-security/code-scanning/automatically-scanning-your-code-for-vulnerabilities-and-errors/configuring-code-scanning#using-queries-in-ql-packs
-        # queries: security-extended,security-and-quality
-
-    # If the analyze step fails for one of the languages you are analyzing with
-    # "We were unable to automatically build your code", modify the matrix above
-    # to set the build mode to "manual" for that language. Then modify this step
-    # to build your code.
-    # ℹ️ Command-line programs to run using the OS shell.
-    # 📚 See https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idstepsrun
-    - name: Run manual build steps
-      if: matrix.build-mode == 'manual'
-      shell: bash
-      run: |
-        echo 'If you are using a "manual" build mode for one or more of the' \
-          'languages you are analyzing, replace this with the commands to build' \
-          'your code, for example:'
-        echo '  make bootstrap'
-        echo '  make release'
-        exit 1
-
-    - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v4
-      with:
-        category: "/language:${{matrix.language}}"
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Initialize CodeQL
+        uses: github/codeql-action/init@v4
+        with:
+          languages: ${{ matrix.language }}
+          build-mode: ${{ matrix.build-mode }}
+
+      - name: Perform CodeQL analysis
+        uses: github/codeql-action/analyze@v4
+        with:
+          category: /language:${{ matrix.language }}
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
deleted file mode 100644
index 22264994..00000000
--- a/.github/workflows/main.yml
+++ /dev/null
@@ -1,42 +0,0 @@
-name: Deploy to Vercel Production
-
-on:
-  push:
-    branches:
-      - main
-  workflow_dispatch:
-
-permissions:
-  contents: read
-
-jobs:
-  deploy:
-    name: Deploy to Vercel
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: pnpm/action-setup@v4
-        with:
-          version: 9
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: '20'
-          cache: pnpm
-
-      - name: Install dependencies
-        run: pnpm install --frozen-lockfile
-
-      - name: Generate Prisma client
-        run: pnpm run db:generate
-        env:
-          POSTGRES_PRISMA_URL: postgresql://ci:ci@localhost/gem
-          POSTGRES_URL_NON_POOLING: postgresql://ci:ci@localhost/gem
-
-      - name: Deploy to Vercel Production
-        run: npx vercel --prod --token $VERCEL_TOKEN --yes
-        env:
-          VERCEL_TOKEN: ${{ secrets.VERCEL_TOKEN }}
-          VERCEL_ORG_ID: ${{ secrets.VERCEL_ORG_ID }}
-          VERCEL_PROJECT_ID: ${{ secrets.VERCEL_PROJECT_ID }}
diff --git a/.vercelignore b/.vercelignore
index 50645e37..31d14123 100644
--- a/.vercelignore
+++ b/.vercelignore
@@ -1,23 +1,14 @@
-# Development and testing artifacts
-src/test/
-*.test.ts
-*.test.tsx
-*.spec.ts
-*.spec.tsx
-
-# Supabase local tooling is not needed at runtime
+# Supabase local tooling is not needed during application builds
 supabase/
 
-# Documentation
+# Documentation is excluded from deployment bundles
 *.md
 DEPLOYMENT.md
 
-# Config files not needed at runtime
+# Local environment templates are not deployed
 .env.example
-eslint.config.mjs
-vitest.config.ts
 
-# Misc
+# Editor and workstation artifacts
 .idea/
 .vscode/
 .DS_Store
diff --git a/AGENTS.md b/AGENTS.md
index b13ac679..42c8344c 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -1,83 +1,142 @@
 # GEM Enterprise — Agent Instructions
 
 ## Project Overview
-GEM Enterprise is a cybersecurity, financial security, asset recovery, and real estate protection platform.
-- **Domain**: gemcybersecurityassist.com
-- **Repository**: github.com/support371/gem-enterprise
-- **Branch**: fix/stabilize-and-finalize-330047221298805619
-- **Deployed on**: Vercel
+
+GEM Enterprise is a controlled-access platform for cybersecurity, compliance, financial-security coordination, product and service enquiries, and property-risk support.
+
+- **Production domain:** `https://www.gemcybersecurityassist.com`
+- **Repository:** `support371/gem-enterprise`
+- **Default branch:** `main`
+- **Canonical host:** Vercel project `support371-gem-enterprise`
+- **Deployment method:** Vercel Git integration deploys `main`; agents must not run a second `vercel --prod` deployment from CI.
+- **Automatic pull-request gate:** canonical Vercel preview builds run lint, type checking, unit tests, and the Next.js production build.
+- **GitHub Actions status:** workflows remain available for manual execution; while hosted-runner access is constrained, their absence is not a passing result and Vercel preview verification is required.
+- **Current operating mode:** controlled production launch. Sensitive or provider-dependent features fail closed until verified.
 
 ## Tech Stack
-- **Framework**: Next.js 16 (App Router)
-- **Language**: TypeScript (strict mode should be enabled)
-- **Database**: PostgreSQL via Prisma 5 ORM
-- **Auth**: Custom JWT using `jose` library + `bcryptjs` for password hashing
-- **Session**: Cookie-based (`gem_session`), httpOnly, 7-day expiry
-- **UI**: Tailwind CSS + shadcn/ui components
-- **Charts**: Recharts
-- **AI**: Anthropic Claude API via direct integration (NOT Vercel AI SDK)
-- **Email**: Resend (when configured)
-- **Payments**: Stripe (when configured)
-- **Package Manager**: pnpm (NEVER use npm or yarn)
-- **Testing**: Vitest (unit) + Playwright (E2E)
-
-## Architecture Rules — DO NOT VIOLATE
-1. **Auth system**: The auth implementation is in `src/lib/auth.ts`. Do NOT replace it with NextAuth, Supabase Auth, Clerk, or any other auth library. Extend the existing JWT implementation.
-2. **Database**: All models go in `prisma/schema.prisma`. Use `db` import from `src/lib/db.ts` for all database access. Never use raw SQL.
-3. **Package manager**: Use `pnpm` for all commands. Never reference npm or yarn.
-4. **API validation**: All API route inputs must be validated with Zod schemas. Never trust raw `req.json()`.
-5. **Audit logging**: All admin and sensitive actions must call `emitAuditLog()` from `src/lib/audit.ts`.
-6. **Evidence retention**: Compliance-sensitive data follows 7-year retention per `src/lib/evidence.ts`.
-7. **Roles**: The system has 5 roles: client, analyst, admin, super_admin, internal. Never allow role escalation to super_admin or internal via API.
-8. **AI consent**: All AI interactions require user consent via `ConsentRecord` model before proceeding.
-9. **File organization**: Pages go in `src/app/`, API routes in `src/app/api/`, shared libraries in `src/lib/`, React components in `src/components/`, hooks in `src/hooks/`.
-10. **Styling**: Match the existing dark theme. Use Tailwind utility classes. Do not add custom CSS files.
-
-## Key Files Reference
-- `src/lib/auth.ts` — JWT sign/verify, session management, KYC-gated routing
-- `src/lib/db.ts` — Prisma client singleton
-- `src/lib/audit.ts` — Audit logging with `emitAuditLog()`
-- `src/lib/evidence.ts` — 7-year evidence retention
-- `src/lib/orchestration/orchestrate-support-reply.ts` — AI chat orchestration
-- `src/lib/policy/evaluate-policy.ts` — AI policy evaluation engine
-- `prisma/schema.prisma` — Database schema (22 models, 9 enums)
-- `next.config.js` — Security headers, redirects, image optimization
-- `vercel.json` — Vercel deployment configuration
-
-## Environment Variables
-Required env vars are documented in `.env.example`. For development/testing, use these defaults:
-- DATABASE_URL=postgresql://test:test@localhost:5432/gem_test
-- JWT_SECRET=test-secret-that-is-at-least-32-characters-long-for-dev
-- NEXT_PUBLIC_APP_URL=http://localhost:3000
-- NEXT_PUBLIC_APP_NAME=GEM Enterprise
-- AUDIT_ENABLED=true
-
-## Verification Commands — RUN AFTER EVERY CHANGE
+
+- **Framework:** Next.js 16 App Router
+- **Language:** TypeScript
+- **Database:** PostgreSQL through Prisma 5
+- **Authentication:** existing JWT/session implementation using `jose` and `bcryptjs`
+- **Session:** HTTP-only cookie named `gem_session`
+- **UI:** Tailwind CSS and existing shared components
+- **Email:** Nodemailer SMTP when production credentials are configured
+- **Commerce:** request-only catalogue until payments, fulfilment, refunds, taxes, licensing, and inventory are verified
+- **Testing:** Vitest; Playwright is available for end-to-end tests
+- **Package manager:** pnpm 10.28.0 only
+
+## Non-Negotiable Safety and Trust Rules
+
+1. **Fail closed.** Do not enable KYC uploads, biometrics, automatic approvals, automatic payments, live intelligence, or marketplace transactions without the complete supporting infrastructure and explicit owner approval.
+2. **No false operational claims.** Do not describe static, sample, mock, planned, or manually reviewed data as live, continuous, verified, encrypted, certified, guaranteed, monitored, or available 24/7.
+3. **No secrets in code, prompts, commits, logs, issues, screenshots, or pull requests.** Use environment-variable names and secret managers only.
+4. **No paid service activation.** An agent may build an adapter or sandbox integration, but may not accept pricing, enable billing, purchase a plan, or create financial obligations.
+5. **No direct production database mutation from CI.** Schema changes require a migration, compatibility assessment, rollback plan, and owner-approved production execution.
+6. **No direct commits to `main`.** Use a focused branch and pull request except for an explicitly documented emergency process.
+7. **One canonical deployment.** Vercel Git integration owns production deployment. Do not create duplicate production deploy workflows or attach the domain to another project without a migration plan.
+8. **Human review for high-impact decisions.** Identity, eligibility, legal, financial, security-incident, and access decisions must remain reviewable and auditable.
+
+## Architecture Rules
+
+1. Extend the authentication implementation in `src/lib/auth.ts`; do not replace it with another auth platform without an approved migration plan.
+2. Put Prisma models in `prisma/schema.prisma` and access the database using `db` from `src/lib/db.ts`.
+3. Validate API inputs with Zod.
+4. Call `emitAuditLog()` from `src/lib/audit.ts` for administrator, reviewer, authorization, verification, and sensitive state changes.
+5. Do not trust client-supplied roles, storage paths, scan results, payment states, provider decisions, or redirect URLs.
+6. Pages belong in `src/app/`, API routes in `src/app/api/`, shared logic in `src/lib/`, and reusable UI in `src/components/`.
+7. Match the existing design system and avoid introducing another styling framework.
+8. External providers must be accessed through small adapter interfaces so they can be disabled or replaced.
+9. Sensitive files must never be stored in public paths or accepted before private storage, validation, scanning, access control, audit logging, retention, and deletion are working.
+10. APIs handling credentials, identity, contact, payments, administration, or recovery must use `no-store` responses and production-appropriate abuse controls.
+
+## Agent Delivery Flow
+
+Every implementation task follows this sequence:
+
+1. **Create or reference one GitHub issue** containing the objective, acceptance criteria, exclusions, risks, and manual dependencies.
+2. **Create a focused branch** using `feat/`, `fix/`, `chore/`, `docs/`, `security/`, or `test/`.
+3. **Inspect before editing.** Read the affected routes, schema, tests, configuration, and recent related pull requests.
+4. **Write a brief implementation plan** in the issue or pull request before broad changes.
+5. **Implement the smallest complete vertical slice.** Avoid unrelated refactors and batch related edits before pushing to reduce unnecessary preview builds.
+6. **Add or update tests** for success, authorization failure, validation failure, provider failure, and fail-closed behavior.
+7. **Run the repository gate locally or in the development agent environment:**
+
 ```bash
+corepack enable
 pnpm install --frozen-lockfile
-pnpm run db:generate
-pnpm build
-pnpm test
+pnpm run verify
 ```
 
-If any of these fail, fix the errors before committing.
-
-## Commit Convention
-- Use conventional commits: feat:, fix:, chore:, test:, docs:
-- Include the prompt number in commits: `feat(prompt-7): wire community hub to database`
-- One logical change per commit when possible
-
-## Testing Standards
-- Unit tests go in `src/__tests__/`
-- E2E tests go in `e2e/`
-- Test files follow pattern: `*.test.ts` or `*.spec.ts`
-- Mock external services (Anthropic, Resend, Stripe) — never call real APIs in tests
-- Target 70%+ statement coverage
-
-## PR Description Format
-When creating a pull request, include:
-1. **Prompt Reference**: Which prompt number this addresses
-2. **Changes Summary**: What was added/modified/deleted
-3. **Files Changed**: List of key files
-4. **Testing**: What tests were added or updated
-5. **Verification**: Confirm `pnpm build` and `pnpm test` pass
+8. **Open a pull request** with risks, manual dependencies, test evidence, and rollback notes.
+9. **Require the canonical Vercel preview to pass.** Preview builds automatically run lint, type checking, unit tests, Prisma generation, and the Next.js production build. Inspect its build logs and affected routes.
+10. **Use the manual GitHub verification and CodeQL workflows when hosted-runner access is available.** Do not describe a workflow that never started as passing.
+11. **Merge only after the available required gates pass.** Vercel Git integration deploys `main` to production.
+12. **Run production smoke checks** without submitting sensitive data or causing billable external actions.
+
+## Build Gate
+
+`pnpm run verify` must complete successfully in the agent environment before the pull request is presented as ready. It runs:
+
+- Prisma client generation
+- ESLint
+- TypeScript checking
+- Unit tests
+- Next.js production build
+
+The canonical Vercel preview independently runs `pnpm run verify:preview` before building the application. This is the automatic hosted gate while GitHub Actions runner access is constrained.
+
+If verification cannot run because a required dependency or account service is unavailable, the agent must document the exact blocker and must not claim the feature is complete.
+
+## Free-Tier Development Policy
+
+Development should preserve useful free operation:
+
+- Prefer existing infrastructure and open-source libraries.
+- Do not add a paid provider merely to complete a demo.
+- Build provider-independent interfaces and use local fakes in tests.
+- Keep optional integrations disabled by default.
+- Batch related commits before pushing so one preview validates one coherent change set.
+- Add usage limits, queue boundaries, retention limits, and clear error states before enabling a metered service.
+- Treat billing approval, identity-provider contracts, production email credentials, regulated data sources, and external seller verification as owner-only actions.
+
+## Current Sensitive Feature Gates
+
+These remain disabled or request-only until their activation requirements pass:
+
+- Identity and financial document upload
+- Biometric and liveness verification
+- Automatic KYC/KYB approval
+- Automatic payments and subscriptions
+- Live threat-intelligence assertions
+- Marketplace transactions and verified-member claims
+- Guaranteed response times and continuous monitoring
+
+## Key Files
+
+- `src/lib/auth.ts` — authentication and session management
+- `src/lib/db.ts` — Prisma client
+- `src/lib/audit.ts` — audit events
+- `src/app/api/contact/route.ts` — durable public enquiry intake
+- `src/app/api/kyc/documents/route.ts` — intentionally fail-closed document endpoint
+- `src/lib/storefrontPresentation.ts` — truthful request-only commerce presentation
+- `prisma/schema.prisma` — database schema
+- `next.config.js` — browser security headers and routing
+- `scripts/vercel-build.mjs` — Prisma, preview verification, and application build orchestration
+- `.github/workflows/ci.yml` — manual full verification fallback
+- `.github/workflows/codeql.yml` — manual CodeQL fallback
+- `docs/AGENT_BUILD_FLOW.md` — task and release operating procedure
+
+## Commit and Pull Request Standards
+
+Use conventional commits such as `feat:`, `fix:`, `security:`, `test:`, `docs:`, and `chore:`. A pull request must include:
+
+1. Objective and linked issue
+2. User-visible and architectural changes
+3. Files and data models affected
+4. Security, privacy, billing, and operational risks
+5. Tests and verification performed
+6. Manual actions that remain
+7. Rollback method
+
+An AI agent is an executor, not the approval authority for billing, legal representations, identity decisions, production secrets, or regulated-service activation.
diff --git a/docs/AGENT_BUILD_FLOW.md b/docs/AGENT_BUILD_FLOW.md
new file mode 100644
index 00000000..61e450ee
--- /dev/null
+++ b/docs/AGENT_BUILD_FLOW.md
@@ -0,0 +1,278 @@
+# GEM Enterprise Agent-First Build Flow
+
+## Purpose
+
+This flow lets GEM Enterprise keep building with limited personnel time and minimal billing exposure while preserving a trustworthy production system.
+
+> Humans define intent and approve high-impact decisions. Agents implement, test, document, and prepare pull requests. Hosted gates verify the work. Sensitive capabilities remain disabled until their infrastructure and owner approvals exist.
+
+## Current hosted-gate arrangement
+
+- **Canonical automatic gate:** Vercel preview deployment for project `support371-gem-enterprise` (`prj_VDGqnA7wZt2E65LLvT94ZOpnYc2Z`).
+- **Duplicate protection:** `scripts/vercel-ignore.mjs` allows only the canonical project to build. Any non-canonical project ID or recognized duplicate URL is ignored before the build begins. Unknown Vercel projects fail closed.
+- **Disconnected duplicates:** `gem-enterprise` (`prj_iT8bNqbTiePiM2SZiWTkOUJXy3o0`) and `project-dtrl6` (`prj_TUZk9mqccnIGSsSpsti7jwg9D3W2`) were disconnected from the repository on 2026-07-09. A fresh commit created no deployments in either project.
+- **What the canonical preview runs:** Prisma generation, ESLint, TypeScript checking, Vitest, and the Next.js production build.
+- **GitHub Actions:** build verification and CodeQL workflows remain available through manual dispatch. They are not automatic while hosted-runner access is constrained.
+- **Production deployment:** Vercel Git integration deploys `main`. No second production deployment workflow is permitted.
+- **Important:** a GitHub workflow that never starts is not a passing check. The canonical Vercel preview and the agent's own `pnpm run verify` evidence are required.
+
+## Roles
+
+### Business owner
+
+Responsible for:
+
+- Business priorities and public claims
+- Billing, contracts, and paid plans
+- Legal entity and policy approval
+- Provider accounts and production credentials
+- Final approval for identity, payment, storage, monitoring, and regulated-service activation
+
+### Lead development agent
+
+Responsible for:
+
+- Repository inspection and task planning
+- Focused implementation
+- Tests and documentation
+- Pull-request creation
+- Local or agent-environment verification
+- Vercel preview verification
+- Reporting exact blockers and manual actions
+
+### Review agent
+
+Responsible for an independent pass over:
+
+- Security and authorization boundaries
+- Privacy and sensitive-data handling
+- False or unsupported public claims
+- Database and migration safety
+- Test coverage and rollback readiness
+
+## Canonical workflow
+
+```text
+Owner objective
+    ↓
+GitHub issue with acceptance criteria
+    ↓
+Agent branch
+    ↓
+Inspect current code and related PRs
+    ↓
+Small vertical implementation
+    ↓
+Unit, authorization, validation, and fail-closed tests
+    ↓
+pnpm run verify in the agent environment
+    ↓
+Pull request
+    ↓
+Duplicate Vercel projects are ignored
+    ↓
+Canonical Vercel preview runs lint + typecheck + tests + build
+    ↓
+Independent review; manual CodeQL when runner access is available
+    ↓
+Owner-only approval when billing, legal, or provider action exists
+    ↓
+Merge to main
+    ↓
+Vercel Git production deployment
+    ↓
+Production smoke check and rollback watch
+```
+
+## Work queue
+
+### P0 — Production safety
+
+Examples:
+
+- Authentication bypass
+- Data exposure
+- Contact submissions being lost
+- Broken production deployment
+- Unsupported claims that can materially mislead users
+
+Rules:
+
+- Stop unrelated feature work.
+- Fix in the smallest possible branch.
+- Require tests and production verification.
+
+### P1 — Revenue and customer flow
+
+Examples:
+
+- Enquiry intake
+- Application status
+- Service-request workflow
+- Email delivery
+- Reviewer queue
+
+Rules:
+
+- Build request-and-review flows before automated or paid flows.
+- Persist data before showing success.
+
+### P2 — Trust infrastructure
+
+Examples:
+
+- Audit records
+- Access controls
+- Backups and restore testing
+- Rate limiting
+- Monitoring
+- Privacy and retention controls
+
+### P3 — Optional expansion
+
+Examples:
+
+- Provider adapters
+- Live feeds
+- Automated payments
+- Biometrics
+- Marketplace transactions
+
+Rule: keep these disabled until P0–P2 requirements and owner approvals are complete.
+
+## Definition of ready
+
+An issue is ready for an agent only when it contains:
+
+- User or business outcome
+- In-scope routes and roles
+- Acceptance criteria
+- Explicit exclusions
+- Sensitive-data classification
+- Billing impact
+- Manual/provider dependencies
+- Rollback expectation
+
+## Definition of done
+
+A task is complete only when:
+
+- The implementation matches the issue scope.
+- Validation and authorization failures are tested.
+- Provider failure or missing configuration fails closed.
+- `pnpm run verify` passes in the development-agent environment.
+- The canonical Vercel preview passes its full preview verification and build.
+- Affected preview routes work.
+- No secret or real sensitive document was used in testing.
+- The PR states what remains manual.
+- Production smoke checks pass after merge.
+
+## Standard agent prompt
+
+```text
+Repository: support371/gem-enterprise
+
+Read AGENTS.md and docs/AGENT_BUILD_FLOW.md before making changes.
+
+Objective:
+<one measurable outcome>
+
+Acceptance criteria:
+- <criterion 1>
+- <criterion 2>
+- <criterion 3>
+
+Explicit exclusions:
+- No paid service activation
+- No production secrets
+- No direct commit to main
+- No enabling sensitive uploads or payments unless every required control and owner approval is included
+
+Required process:
+1. Inspect affected code, schema, tests, configuration, and recent related PRs.
+2. Post a short plan.
+3. Implement the smallest complete vertical slice.
+4. Add tests for success, validation failure, authorization failure, and fail-closed behavior.
+5. Run pnpm run verify in your environment.
+6. Batch related changes before pushing to reduce preview builds.
+7. Open a PR with risks, manual dependencies, and rollback steps.
+8. Require the canonical Vercel preview to pass.
+9. Do not claim completion if any available gate fails.
+```
+
+## Recommended agent allocation
+
+### Primary implementation agent: OpenAI Codex
+
+Use for:
+
+- Multi-file repository tasks
+- Bug diagnosis
+- Refactors and migrations
+- Test creation
+- Pull-request preparation
+- Parallel tasks isolated by branch
+
+Codex must follow `AGENTS.md`, use one issue per task, run repository verification in its environment, and operate through pull requests.
+
+### Independent review: GitHub Copilot coding agent or a second Codex task
+
+Use for:
+
+- Reviewing the PR without sharing the implementation conversation
+- Checking authorization, data handling, and regressions
+- Finding missing tests
+
+### Replit Agent
+
+Use for a visual prototype or isolated experiment. Do not make it the canonical production repository or deployment path.
+
+### Devin
+
+Use only for a clearly bounded backlog when paid capacity is approved. It is not required for the current free-first flow.
+
+## First implementation program: GEM Verify Core
+
+Build in this order:
+
+1. Verification application and consent records
+2. Manual case-status workflow
+3. Reviewer queue and role checks
+4. Information-request, approval, rejection, and escalation actions
+5. Audit events and immutable decision history
+6. Storage-provider interface with a fake test implementation
+7. Retention and deletion scheduling model
+8. Private upload adapter only after owner-selected storage and scanning are available
+9. Optional Veriff or another provider adapter after sandbox approval
+
+Until step 8 is verified, `/api/kyc/documents` must remain fail closed.
+
+## Free-tier safeguards
+
+- Batch related edits before pushing so one preview validates one coherent change set.
+- Cancel superseded preview and CI runs where supported.
+- Use the canonical Vercel project only.
+- Let Vercel Git integration perform the only production deployment.
+- Keep `scripts/vercel-ignore.mjs` fail closed for any unknown or non-canonical Vercel project.
+- Use local fakes for external providers in tests.
+- Do not run scheduled jobs more frequently than required.
+- Keep optional integrations disabled by default.
+- Record usage and define a hard stop before enabling metered services.
+- Never rely on a free plan whose terms prohibit the intended business use.
+
+## Manual gates
+
+The agent must stop and request owner action for:
+
+- Accepting pricing or terms
+- Enabling billing
+- Creating regulated-data provider accounts
+- Entering production secrets
+- Selecting retention periods
+- Approving public certifications or SLAs
+- Naming authorized identity reviewers
+- Activating payment collection
+- Moving the production domain or database
+- Disconnecting duplicate Vercel projects or changing organization-level GitHub billing and Actions settings
+
+These gates are separation-of-duty controls, not development failures.
diff --git a/docs/VERCEL_CANONICAL_PROJECT.md b/docs/VERCEL_CANONICAL_PROJECT.md
new file mode 100644
index 00000000..8ad4b241
--- /dev/null
+++ b/docs/VERCEL_CANONICAL_PROJECT.md
@@ -0,0 +1,41 @@
+# Canonical Vercel Project
+
+Verified on 2026-07-09 after the dashboard cleanup.
+
+## Canonical project
+
+- Name: `support371-gem-enterprise`
+- Project ID: `prj_VDGqnA7wZt2E65LLvT94ZOpnYc2Z`
+- Production domains:
+  - `https://www.gemcybersecurityassist.com`
+  - `https://gemcybersecurityassist.com`
+
+This is the only Vercel project authorized to build and deploy `support371/gem-enterprise`.
+
+## Disconnected duplicate projects
+
+- `gem-enterprise` — `prj_iT8bNqbTiePiM2SZiWTkOUJXy3o0`
+- `project-dtrl6` — `prj_TUZk9mqccnIGSsSpsti7jwg9D3W2`
+
+After their Git connections were removed, neither duplicate created a deployment from subsequent repository activity. The repository-level `scripts/vercel-ignore.mjs` guard remains enabled as defense in depth and fails closed for non-canonical or unknown Vercel projects.
+
+## Verified preview gate
+
+Canonical preview deployment `dpl_7zyuXQAuMaVE1ywj6eo4HRoyS2C6` reached `READY` for code commit `41050e0c08449c6bc282c4ead34c4348ca3330a1` after completing:
+
+- Prisma client generation
+- ESLint
+- TypeScript checking
+- 14 Vitest files and 164 passing tests
+- Next.js 16 production compilation and deployment
+
+This file is documentation-only and excluded from the deployment bundle by `.vercelignore`, so later documentation commits do not change the verified application build.
+
+## Verification rule
+
+Before merging a pull request:
+
+1. Confirm the pull-request code head creates a deployment only in the canonical project.
+2. Confirm the canonical preview reaches `READY`.
+3. Inspect the canonical build logs for lint, TypeScript, unit-test, Prisma, and Next.js build results.
+4. Do not treat historical checks from disconnected projects as current deployment evidence.
diff --git a/package.json b/package.json
index 63916664..a3fc9a08 100644
--- a/package.json
+++ b/package.json
@@ -2,14 +2,18 @@
   "name": "gem-enterprise",
   "private": true,
   "version": "1.0.0",
+  "packageManager": "pnpm@10.28.0",
   "scripts": {
     "dev": "next dev",
     "build": "node scripts/vercel-build.mjs",
     "start": "next start",
     "lint": "eslint src --ext .ts,.tsx --max-warnings=0",
+    "typecheck": "tsc --noEmit",
     "test": "vitest run",
     "test:watch": "vitest",
     "test:coverage": "vitest run --coverage",
+    "verify:preview": "pnpm run lint && pnpm run typecheck && pnpm run test",
+    "verify": "pnpm run db:generate && pnpm run verify:preview && pnpm run build",
     "db:generate": "prisma generate",
     "db:push": "prisma db push",
     "db:migrate": "prisma migrate dev",
@@ -117,4 +121,4 @@
     "typescript": "^5.8.3",
     "vitest": "^4.1.0"
   }
-}
\ No newline at end of file
+}
diff --git a/scripts/vercel-build.mjs b/scripts/vercel-build.mjs
index 21e6d07d..67ba5619 100644
--- a/scripts/vercel-build.mjs
+++ b/scripts/vercel-build.mjs
@@ -39,6 +39,27 @@ if (directUrl) env.POSTGRES_URL_NON_POOLING = directUrl;
 console.log("Generating Prisma client...");
 run("pnpm", ["exec", "prisma", "generate"], env);
 
+const shouldVerifyPreview =
+  env.VERCEL_ENV === "preview" || env.RUN_PREVIEW_VERIFICATION === "true";
+
+if (shouldVerifyPreview) {
+  console.log("Running preview verification: lint, typecheck, and unit tests...");
+  const verificationEnv = {
+    ...env,
+    NODE_ENV: "test",
+    JWT_SECRET: "preview-verification-secret-min-32-characters",
+    POSTGRES_PRISMA_URL: "postgresql://ci:ci@localhost:5432/gem_ci",
+    POSTGRES_URL_NON_POOLING: "postgresql://ci:ci@localhost:5432/gem_ci",
+    SMTP_HOST: "",
+    SMTP_PORT: "587",
+    SMTP_USER: "",
+    SMTP_PASS: "",
+    CRON_SECRET: "preview-verification-cron-secret",
+    AUDIT_ENABLED: "true",
+  };
+  run("pnpm", ["run", "verify:preview"], verificationEnv);
+}
+
 if (env.AUTO_DB_PUSH === "true") {
   if (!pooledUrl) {
     throw new Error(
diff --git a/scripts/vercel-ignore.mjs b/scripts/vercel-ignore.mjs
index c42ca396..4a9012c1 100644
--- a/scripts/vercel-ignore.mjs
+++ b/scripts/vercel-ignore.mjs
@@ -1,15 +1,66 @@
+// Defense-in-depth: only the canonical GEM Vercel project may consume a build.
 const canonicalProjectId = "prj_VDGqnA7wZt2E65LLvT94ZOpnYc2Z";
-const currentProjectId = process.env.VERCEL_PROJECT_ID?.trim();
+const canonicalMarkers = [
+  "support371-gem-enterprise",
+  "www.gemcybersecurityassist.com",
+  "gemcybersecurityassist.com",
+];
+const knownDuplicateMarkers = [
+  "project-dtrl6",
+  "gem-enterprise-admin-25521151s-projects",
+  "gem-enterprise-git-",
+];
 
-if (!currentProjectId) {
-  console.log("VERCEL_PROJECT_ID is unavailable; continue the build safely.");
+const projectId = process.env.VERCEL_PROJECT_ID?.trim() || "";
+const deploymentSignals = [
+  process.env.VERCEL_PROJECT_PRODUCTION_URL,
+  process.env.VERCEL_URL,
+  process.env.VERCEL_BRANCH_URL,
+  process.env.VERCEL_TARGET_ENV,
+]
+  .filter((value) => typeof value === "string" && value.trim().length > 0)
+  .map((value) => value.trim().toLowerCase());
+
+function continueBuild(reason) {
+  console.log(`[vercel-ignore] Continue build: ${reason}`);
   process.exit(1);
 }
 
-if (currentProjectId === canonicalProjectId) {
-  console.log(`Canonical Vercel project ${canonicalProjectId}; continue the build.`);
-  process.exit(1);
+function ignoreBuild(reason) {
+  console.log(`[vercel-ignore] Ignore build: ${reason}`);
+  process.exit(0);
+}
+
+if (projectId) {
+  if (projectId === canonicalProjectId) {
+    continueBuild(`canonical project ID ${canonicalProjectId}`);
+  }
+
+  ignoreBuild(
+    `non-canonical project ID ${projectId}; canonical project is ${canonicalProjectId}`,
+  );
+}
+
+if (
+  deploymentSignals.some((signal) =>
+    canonicalMarkers.some((marker) => signal.includes(marker)),
+  )
+) {
+  continueBuild(`canonical project marker detected in ${deploymentSignals.join(", ")}`);
+}
+
+if (
+  deploymentSignals.some((signal) =>
+    knownDuplicateMarkers.some((marker) => signal.includes(marker)),
+  )
+) {
+  ignoreBuild(`known duplicate marker detected in ${deploymentSignals.join(", ")}`);
+}
+
+if (process.env.VERCEL === "1" || process.env.CI === "1") {
+  ignoreBuild(
+    "Vercel project identity could not be verified. Failing closed to prevent an unknown or duplicate project from consuming build capacity.",
+  );
 }
 
-console.log(`Ignoring duplicate Vercel project ${currentProjectId}; canonical project is ${canonicalProjectId}.`);
-process.exit(0);
+continueBuild("local non-Vercel environment");