@sun-asterisk/sungen 2.4.6 → 2.5.1

package/src/orchestrator/templates/ai-instructions/github-skill-sungen-gherkin-syntax.md

@@ -151,6 +151,14 @@ Options: `nth` `exact` `scope` `match` `variant` `frame` `contenteditable` `colu
 | `@no-auth` | Disable inherited auth |
 | `@steps:name` | Define reusable step block (base scenario) |
 | `@extend:name` | Prepend Given→When from @steps block (skip Then) |
+| `@cleanup:overlay` | Auto-cleanup: dismiss dialogs/overlays after each test (base.ts fixture) |
+| `@cleanup:forms` | Auto-cleanup: clear form fields after each test (base.ts fixture) |
+| `@cleanup:scroll` | Auto-cleanup: scroll to top after each test (base.ts fixture) |
+| `@cleanup:storage` | Auto-cleanup: clear sessionStorage after each test (base.ts fixture) |
+| `@screenshot:on-failure` | Auto-capture screenshot when test fails (base.ts fixture) |
+| `@beforeAll` | Hook: runs once before all tests → `test.beforeAll()` |
+| `@afterEach` | Hook: runs after each test → `test.afterEach()` (custom cleanup) |
+| `@afterAll` | Hook: runs once after all tests → `test.afterAll()` |
 
 ### @extend behavior
 
@@ -175,30 +183,28 @@ Options: `nth` `exact` `scope` `match` `variant` `frame` `contenteditable` `colu
 | Missing `is` for state | `with {{text}} hidden` | `with {{text}} is hidden` |
 | State as value | `with {{disabled}}` | `is disabled` |
 | Missing target type | `fill [email] with {{v}}` | `fill [email] field with {{v}}` |
-| Using Background | `Background: Given User is on...` | Use `@steps` + `@extend` instead |
+| Background with scope | `Background: ... And User is on [X] dialog` | Use `@steps` + `@extend` for scope-dependent flows |
 | `is on` after When | `When ... And User is on [X] dialog` | `And User see [X] dialog` or separate Given |
 
 ## Background vs @steps/@extend
 
-**Do NOT use `Background:` block.** Use `@steps` + `@extend` instead.
+Both `Background` and `@steps`/`@extend` are valid — they serve different purposes.
 
-**Why:**
-- `Background` runs before EVERY scenario but cannot set dialog/frame scope correctly
-- `Background` with `When` + `And User is on [X] dialog` creates keyword mismatch (`is on` = Given, not When)
-- `@steps` + `@extend` gives explicit control: base steps run Given→When, extending scenario starts with `Given User is on [X] dialog` (correct scope setup)
+| Pattern | Use when | Generates |
+|---|---|---|
+| `Background` | Simple shared setup (navigate to page) | `test.beforeEach()` |
+| `@steps`/`@extend` | Complex reusable flows with scope (dialog, frame) | Inline merged steps in `test()` |
 
-**Wrong:**
+**Use `Background` for simple navigation:**
 ```gherkin
 Background:
-  Given User is on [Kudos] page
-  When User click [Open] button
-  And User is on [Modal] dialog    ← keyword mismatch
+  Given User is on [Dashboard] page
 
-Scenario: VP-UI-001 Title visible
-  Then User see [Title] heading
+Scenario: View stats
+  Then User see [Revenue Chart] section
 ```
 
-**Correct:**
+**Use `@steps`/`@extend` when scope matters (dialog, frame):**
 ```gherkin
 @steps:open_modal
 Scenario: Open modal
@@ -211,3 +217,70 @@ Scenario: VP-UI-001 Title visible
   Given User is on [Modal] dialog
   Then User see [Title] heading
 ```
+
+**Avoid `Background` with scope-dependent steps** — `When` + `And User is on [X] dialog` creates keyword mismatch (`is on` = Given, not When). Use `@steps`/`@extend` instead.
+
+## Hooks & Cleanup
+
+Two layers for test lifecycle management. Prefer `@cleanup:*` tags (Layer 1) — they work with base.ts automatically. Use hook scenarios (Layer 2) only for custom logic.
+
+### Layer 1: `@cleanup:*` tags (automatic via base.ts)
+
+Feature-level tags that activate cleanup fixtures in base.ts. No Gherkin steps needed.
+
+```gherkin
+@auth:admin
+@cleanup:overlay
+@cleanup:forms
+Feature: User Management
+  Path: /users
+
+  Background:
+    Given User is on [User Management] page
+
+  Scenario: Create user shows form
+    When User click [Add User] button
+    Then User see [Create User] dialog
+    # After test: overlay auto-dismissed, forms auto-cleared by base.ts
+
+  Scenario: Search user by name
+    When User fill [Search] field with {{search_name}}
+    Then User see [User Row] row
+    # After test: search field auto-cleared by base.ts
+```
+
+| Tag | What base.ts does after each test |
+|---|---|
+| `@cleanup:overlay` | Press Escape, click body, dismiss fixed overlays |
+| `@cleanup:forms` | Clear all input/textarea fields, reset selects |
+| `@cleanup:scroll` | Scroll to top of page |
+| `@cleanup:storage` | Clear sessionStorage |
+
+### Layer 2: `@afterEach` scenario (custom cleanup)
+
+Only when `@cleanup:*` tags aren't enough — feature-specific logic.
+
+```gherkin
+@auth:admin
+@cleanup:overlay
+Feature: Dashboard
+  Path: /dashboard
+
+  Background:
+    Given User is on [Dashboard] page
+
+  @afterEach
+  Scenario: Reset dashboard filters
+    When User select [Date Filter] dropdown with {{default_period}}
+
+  Scenario: Filter by last week
+    When User select [Date Filter] dropdown with {{last_week}}
+    Then User see [Revenue Chart] section
+```
+
+### Layer 3: `@beforeAll` / `@afterAll` (optional)
+
+For one-time setup/teardown. Low priority — most e2e tests don't need these.
+
+**Rendering order in `.spec.ts`:**
+`test.describe` → `test.use(storageState)` → `test.use(autoCleanup)` → `test.beforeAll` → `test.beforeEach` → `test.afterEach` → `test.afterAll` → `test()` blocks

package/src/orchestrator/templates/ai-instructions/github-skill-sungen-selector-fix.md

@@ -52,6 +52,67 @@ If existing selectors already cover the feature file, **skip Phase 0** and go st
 
 ---
 
+## Phase 0 (Provisional): Figma-Only Selector Generation (no live page)
+
+**Trigger**: Phase 0 decision tree (in `run-test` command) determined the live page is not reachable AND `requirements/spec_figma.md` exists.
+
+**Goal**: produce a compilable `selectors.yaml` seeded from Figma node data so that `sungen generate` succeeds and the test suite can run. Every entry is provisional — it must be verified against a real DOM when the page becomes available.
+
+### Inputs
+
+- `requirements/spec_figma.md` — specifically:
+  - `## Components` table (Name, Type, Role, Text, Variants columns)
+  - `## Text Inventory` list (node_id → label text pairs)
+
+### Steps
+
+1. **Parse references**: collect every `[Reference]` element from the `.feature` file (same as live Phase 0 step 2).
+2. **Match to Figma nodes**: for each `[Reference]`, find the best matching entry in `spec_figma.md` by comparing the reference name against component names and text labels (case-insensitive, partial match allowed).
+3. **Apply heuristic priority** (same order as live Phase 0 — use first applicable):
+
+   | Priority | Type | Figma Signal |
+   |---|---|---|
+   | 1 | `testid` | Dev annotation `data-testid=X` in component metadata |
+   | 2 | `role` + name | Component type is Button/Link + has text label |
+   | 3 | `placeholder` | Component type is Input/Field + has placeholder text |
+   | 4 | `label` | Component has associated label text (not placeholder) |
+   | 5 | `locator` (CSS) | No accessible name derivable from Figma data |
+   | 6 | `text` | Plain text node (not inside interactive component) |
+
+   See `sungen-figma-source` skill for the authoritative heuristics table mapping Figma signals to YAML entry types.
+
+4. **Write entries**: add each entry to `selectors.yaml`, prefixed with the required comment:
+   ```yaml
+   # @needs-live-verify source=figma node_id=<figma-node-id>
+   submit-button:
+     type: role
+     value: button
+     name: "Submit"
+   ```
+   - Key naming follows `sungen-selector-keys` conventions (lowercase, Unicode preserved, `--type` / `--N` suffixes).
+   - Copy text character-for-character from `## Text Inventory`. Never infer from the Gherkin label.
+   - Merge, don't overwrite — preserve the page selector and any user-authored entries.
+
+5. **Compile**: `sungen generate --screen <screen>` — must succeed before Phase 1.
+
+### Auto-fix on live page (subsequent run-test invocations)
+
+When `run-test` is called again and the live page is now reachable:
+- Phase 0 takes ONE `browser_snapshot`.
+- For each entry tagged `# @needs-live-verify` in `selectors.yaml`:
+  - Compare the provisional selector against the actual DOM element found in the snapshot.
+  - If match → remove the comment line (entry promoted to verified).
+  - If mismatch → replace the YAML value with the snapshot-derived selector, then remove the comment (entry corrected and verified).
+- Entries with no corresponding DOM element → log as unresolved, leave comment in place for manual inspection.
+
+### Pitfalls
+
+- Inferring selector names from Gherkin labels instead of Figma `## Text Inventory` → mismatch on live verification.
+- Omitting `# @needs-live-verify` comment → reviewer has no way to count unverified selectors.
+- Overwriting user-authored selectors during merge → always check for existing keys before adding.
+
+---
+
 ## Phase 0.5: Auth Persistence (MCP alternative to `sungen makeauth`)
 
 Capture an authenticated session from the MCP browser into `specs/.auth/<role>.json` — the same path `sungen makeauth` writes to, which compiled tests already reference via `test.use({ storageState })` based on `@auth:<role>` tags. No `playwright.config.ts` edits needed. Run once per auth lifetime, not on every selector fix.

package/src/orchestrator/templates/ai-instructions/github-skill-sungen-tc-generation.md

@@ -6,14 +6,25 @@ user-invocable: false
 
 ## Goal
 
-Generate **focused test cases per screen section** with **20+ scenarios per viewpoint**. Output `.feature` + `test-data.yaml` only — selectors are deferred to `/sungen:run-test`.
+Generate **focused test cases per screen section** using a **tier-based approach** for faster results. Output `.feature` + `test-data.yaml` only — selectors are deferred to `/sungen:run-test`.
+
+### Tier System
+
+| Tier | Priority | What to generate | When |
+|---|---|---|---|
+| **Tier 1** (default) | `@critical` + `@high` | Happy paths, required validation, core business rules, security basics | First run of `create-test` |
+| **Tier 2** (expand) | `@normal` + `@low` | UI presence, optional validation, edge cases, cosmetic checks | User runs `create-test` again with "Add viewpoints" mode |
+
+**Round 1 (Tier 1)** targets **~10-15 scenarios per section** — enough to cover critical flows and catch real bugs. This is the default behavior.
+
+**Round 2 (Tier 2)** expands to full coverage when the user explicitly chooses "Add viewpoints" or "Add new sections" update mode. Only then generate `@normal` + `@low` scenarios to fill coverage gaps.
 
 ## Update Mode
 
 When `.feature` already has scenarios, summarize and ask:
-1. **Add new sections** — append, continue numbering
-2. **Add viewpoints** — add to existing sections
-3. **Replace all** — overwrite
+1. **Add new sections** — append new sections with Tier 2 (`@normal` + `@low`) scenarios, continue numbering
+2. **Add viewpoints** — expand existing sections with Tier 2 (`@normal` + `@low`) scenarios
+3. **Replace all** — overwrite with fresh Tier 1 (`@critical` + `@high`) generation
 
 For append: read highest `VP-<CAT>-<NNN>`, continue from next number. Never modify existing scenarios.
 
@@ -28,14 +39,28 @@ Requirements improve every viewpoint: exact error messages for VAL, business rul
 
 If also exploring live page: verify spec vs actual, flag mismatches, capture exact text.
 
+### Figma supplement (`spec_figma.md`)
+
+When `requirements/spec_figma.md` is present alongside `spec.md`, treat it as a **secondary input** with these rules:
+
+- **Never override `spec.md`**: `spec.md` is authoritative for all business rules, field constraints, and behavior. `spec_figma.md` only supplements with visual/text data that `spec.md` may lack.
+- **`## Text Inventory` → literal strings**: use text label values from this section verbatim in `test-data.yaml` (button labels, input placeholders, error messages shown in Figma). Do not paraphrase or invent alternatives.
+- **`## Interaction States` → state coverage checkpoints**: use the listed variants (e.g., empty, loading, error, success) as a checklist for state-coverage scenarios. Only generate scenarios for states that are either (a) confirmed in both `spec.md` and `spec_figma.md`, or (b) explicitly documented in one source without contradiction from the other.
+- **Flag disagreements**: if a field name, label, or behavior in `spec_figma.md` contradicts `spec.md`, insert an HTML comment at the top of the `.feature` file:
+  ```gherkin
+  <!-- FIGMA-SPEC CONFLICT: <brief description> — using spec.md value -->
+  ```
+  Then proceed using the `spec.md` value.
+
 ## Screen Input Sources
 
-**Recommended** (in priority order):
-1. **Figma designs** — use Figma MCP to read design context, screenshots, and component metadata
-2. **UI images** — screenshots or mockups in `qa/screens/<screen>/requirements/ui/`
-3. **`spec.md`** — written specification with sections, fields, validation rules
+**Auto-detect** — the parent command (`create-test`) resolves visual sources before invoking this skill. By the time generation starts, the available sources are already determined:
+- `spec.md` — primary, always read if present
+- `spec_figma.md` — Figma supplement, read if present (PAT flow already completed)
+- `ui/*.png` — visual context, read if present
+- `test-viewpoint.md` — edge cases and known issues, read if present
 
-**Optional**: **Live page scan** via Playwright MCP — useful to verify specs vs actual UI, capture real data (error messages, labels). If auth needed → ask user to log in manually via MCP browser.
+**IMPORTANT:** If `spec_figma.md` exists, do NOT call any `mcp__figma__*` tool. The PAT flow is complete — just read the file.
 
 **Single screen focus**: one URL = one screen. Don't explore sibling paths. Modals on same page = part of this screen.
 
@@ -76,7 +101,9 @@ Apply `sungen-test-design-techniques` to spec-extracted conditions:
 
 Use `sungen-viewpoint` skill for per-pattern checklists across 4 viewpoints: UI/UX, Data & Validate, Logic, Security.
 
-Add scenarios for generic UI coverage that spec didn't explicitly state (empty states, loading states, keyboard nav, hover effects). Skip viewpoints truly N/A.
+**Tier-aware gap filling:**
+- **Tier 1 (first run)**: only add `@critical` and `@high` items from the checklists — core security checks (VP-SEC), required field validation (VP-VAL), key state transitions (VP-LOGIC). Skip `@normal`/`@low` items like hover states, empty states, tooltips.
+- **Tier 2 (expand run)**: add `@normal` + `@low` scenarios — UI presence, optional validation, edge cases, cosmetic checks, keyboard nav, hover effects.
 
 **Validation rule**: capture actual error messages from live page or spec.md. Use `User see {{error_var}}` — never assert just "is visible".
 
@@ -112,30 +139,70 @@ Given User is on [Screen] page
 And User wait for [Page Title] heading is visible
 ```
 
+## Cleanup & Hooks
+
+### Auto-assign `@cleanup:*` tags based on screen sections
+
+After identifying screen sections, add appropriate `@cleanup:*` feature-level tags. These activate base.ts fixtures that auto-clean state between tests.
+
+| Screen has | Add tag | Why |
+|---|---|---|
+| Modal / Dialog / Drawer | `@cleanup:overlay` | Dismiss leftover overlays between tests |
+| Form & Inputs / Search / Filter | `@cleanup:forms` | Clear form fields, reset selects |
+| Long scrollable content | `@cleanup:scroll` | Scroll to top for consistent assertions |
+| Auth tokens / session data in tests | `@cleanup:storage` | Clear sessionStorage |
+| CI/CD or debug-heavy screens | `@screenshot:on-failure` | Auto-capture screenshot on test failure |
+
+**Always add `@cleanup:overlay`** if ANY section opens a dialog (Create/Add, Update/Edit, Delete confirmation). Most CRUD screens need it.
+
+**Always add `@cleanup:forms`** if the screen has inline search, filter dropdowns, or editable forms that persist between tests.
+
+### When to add `@afterEach` hook scenario
+
+Only when `@cleanup:*` tags aren't enough — feature-specific cleanup logic:
+- Reset a dropdown filter to default value (not just clear)
+- Navigate away from a sub-tab back to the main tab
+- Close a specific sidebar panel
+
+```gherkin
+@afterEach
+Scenario: Reset filters to default
+  When User select [Status Filter] dropdown with {{default_status}}
+```
+
+### `@beforeAll` / `@afterAll` — optional, low priority
+
+For one-time setup/teardown. Most screens don't need these.
+
 ## Output Format
 
 **Feature file** — `qa/screens/<screen>/features/<screen>.feature`
 
-**Never use `Background:`.** Use `@steps` + `@extend` for shared setup (see `sungen-gherkin-syntax` skill).
+`Background` is valid for simple shared setup (navigate to page). Use `@steps`/`@extend` for complex flows with scope (dialog, frame).
 
 ```gherkin
 @auth:role
+@cleanup:overlay
+@cleanup:forms
 Feature: <Screen> Screen
 
+  Background:
+    Given User is on [Screen] page
+
   # Shared setup — NO priority tag on @steps
   @steps:open_form
   Scenario: Open form
-    Given User is on [Screen] page
     When User click [Create] button
     Then User see [Form] dialog
 
-  # --- Section: Create User Form ---
+  # --- Section: Create User Form (Tier 1: @critical + @high) ---
 
-  @normal @extend:open_form
-  Scenario: VP-UI-001 Form displays all fields with correct defaults
+  @critical @extend:open_form
+  Scenario: VP-LOGIC-001 Submit form with valid data creates record
     Given User is on [Form] dialog
-    Then User see [Name] field
-    And User see [Submit] button is disabled
+    When User fill [Name] field with {{valid_name}}
+    And User click [Submit] button
+    Then User see {{success_message}} message
 
   @high @extend:open_form
   Scenario: VP-VAL-001 Submit with all empty fields shows errors
@@ -143,11 +210,7 @@ Feature: <Screen> Screen
     When User click [Submit] button
     Then User see [Name error] message with {{name_required_error}}
 
-  # --- Section: User Table ---
-
-  @normal
-  Scenario: VP-UI-010 Table displays all columns
-    Then User see [Name] column in [Users] table
+  # --- Section: User Table (Tier 1: @critical + @high) ---
 
   @high
   Scenario: VP-VAL-010 Table displays correct data
@@ -155,13 +218,15 @@ Feature: <Screen> Screen
       | Name       | Email        |
       | {{name_1}} | {{email_1}}  |
 
-  @high
-  Scenario: VP-VAL-011 Edit button targets correct row
-    Given User see [Target] row in [Users] table with {{name_1}}
-    When User click [Edit] button in [Users] table with {{name_1}}
-    Then User see [Name] field with {{name_1}}
+  @critical
+  Scenario: VP-SEC-010 Unauthorized user cannot access page
+    Given User is not logged in
+    When User navigate to [Screen] page
+    Then User is on [Login] page
 ```
 
+**Tier 2 (expand run)** adds `@normal` + `@low` scenarios like UI field presence, hover states, tooltips, empty states.
+
 ### When to use DataTable vs Row Scope
 
 | Pattern | Use when |
@@ -171,6 +236,18 @@ Feature: <Screen> Screen
 
 **Naming**: `VP-<CATEGORY>-<NNN>` prefix. Scenario name must use the **same element type** as the steps — e.g., if the step uses `dialog`, write "dialog opens" not "modal opens".
 
-**Test data** — `qa/screens/<screen>/test-data/<screen>.yaml`, grouped by section.
+**Test data** — `qa/screens/<screen>/test-data/<screen>.yaml`, grouped by section. Data is loaded **at runtime** — keys become runtime lookups, not hardcoded strings. The same compiled test works across environments.
+
+**Environment-specific data**: For values that differ per environment (credentials, URLs, test users), create `<screen>.<env>.yaml` alongside the base file. Users run `SUNGEN_ENV=staging npx playwright test` to merge overrides. Structure env YAML with the same keys, only including values that change:
+
+```yaml
+# login.yaml (base)
+valid_email: admin@dev.example.com
+valid_password: DevPass123
+
+# login.staging.yaml (override for staging)
+valid_email: admin@staging.example.com
+valid_password: StagingPass456
+```
 
 **Do NOT generate**: `selectors.yaml` (created during run-test), Playwright code (sungen compiles).

package/src/orchestrator/templates/ai-instructions/github-skill-sungen-tc-review.md

@@ -43,6 +43,8 @@ Score: `(dimensions_covered / 6) * 40`. Validate technique application with `sun
 
 **Classification**: UI = static/always-same appearance. VAL = input validation/errors. LOGIC = behavior/state changes (includes persisted state without When). SEC = auth/permissions.
 
+**Tier-aware scoring**: If the feature file only contains `@critical` + `@high` scenarios (Tier 1), do NOT penalize for missing VP-UI viewpoint — UI scenarios are intentionally deferred to Tier 2. Score "All applicable VP present" based on Tier 1-relevant viewpoints only (VAL, LOGIC, SEC). Note in the review output: *"VP-UI deferred to Tier 2 — run create-test with 'Add viewpoints' to expand."*
+
 ---
 
 ## Quality Rules
@@ -87,6 +89,21 @@ Do NOT mark `@manual` when data is visible in snapshot or documented in spec —
 
 ---
 
+## Unverified Selectors metric
+
+**When to check**: if `qa/screens/<screen>/selectors/<screen>.yaml` exists, count lines matching the pattern `@needs-live-verify`.
+
+| Metric | Source | Scoring impact |
+|---|---|---|
+| Unverified Selectors | Count of `@needs-live-verify` comment lines in `selectors.yaml` | None — non-blocking |
+
+- If count > 0: include the following line in the review report summary (after the score table, before Issues):
+  `⚠ <N> selectors flagged @needs-live-verify — run run-test against live URL when available.`
+- Does NOT reduce the score or block the 60% threshold. The suite can PASS with unverified selectors.
+- If `selectors.yaml` does not exist yet (not yet generated): omit this metric entirely.
+
+---
+
 ## Output Format
 
 ```markdown
@@ -99,6 +116,9 @@ Do NOT mark `@manual` when data is visible in snapshot or documented in spec —
 | Viewpoint | <n> | 30 |
 | **Total** | **<n>%** | **100** |
 
+⚠ <N> selectors flagged @needs-live-verify — run run-test against live URL when available.
+<!-- omit this line if selectors.yaml doesn't exist or @needs-live-verify count = 0 -->
+
 ### Issues
 1. [SYNTAX] ...
 2. [COVERAGE] ...

package/src/orchestrator/templates/specs-base.ts

@@ -1,12 +1,27 @@
 import { test as base, expect } from '@playwright/test';
 import type { BrowserContext, Page } from '@playwright/test';
 
+type CleanupConfig = {
+  overlay?: boolean;
+  forms?: boolean;
+  scroll?: boolean;
+  storage?: boolean;
+};
+
 // Share one context per storageState — avoids creating multiple sessions
 // that trigger server rate limiting or session invalidation
 const contextCache = new Map<string, { context: BrowserContext; page: Page }>();
 const GOTO_PATCHED = Symbol('goto-patched');
 
-const test = base.extend({
+const test = base.extend<{
+  autoCleanup: CleanupConfig;
+  screenshotOnFailure: boolean;
+  _autoCleanup: void;
+  _autoScreenshot: void;
+}>({
+  autoCleanup: [{}, { option: true }],
+  screenshotOnFailure: [false, { option: true }],
+
   page: async ({ browser, storageState }, use) => {
     if (storageState) {
       const cacheKey = typeof storageState === 'string' ? storageState : JSON.stringify(storageState);
@@ -28,12 +43,6 @@ const test = base.extend({
           try {
             const currentPath = new URL(page.url()).pathname;
             if (currentPath === url || currentPath === url + '/') {
-              // Dismiss any open overlays (dropdowns, dialogs) from previous test
-              await page.keyboard.press('Escape').catch(() => {});
-              await page.locator('body').click({ position: { x: 1, y: 1 }, force: true }).catch(() => {});
-
-              // Safety check: if a fixed-position overlay (modal/dialog) is still present, full reload
-              // eslint-disable-next-line no-eval -- runs in browser context via Playwright
               const hasOverlay = await page.evaluate(`(() => {
                 const el = document.elementFromPoint(window.innerWidth / 2, window.innerHeight / 2);
                 if (!el) return false;
@@ -68,6 +77,55 @@ const test = base.extend({
       await context.close();
     }
   },
+
+  // Auto-cleanup fixture: runs teardown after each test based on @cleanup:* tags
+  _autoCleanup: [async ({ page, autoCleanup }, use) => {
+    await use();
+
+    if (autoCleanup.overlay) {
+      await page.keyboard.press('Escape').catch(() => {});
+      await page.locator('body').click({ position: { x: 1, y: 1 }, force: true }).catch(() => {});
+      // Dismiss persistent fixed overlays (modals, dialogs)
+      const hasOverlay = await page.evaluate(`(() => {
+        const el = document.elementFromPoint(window.innerWidth / 2, window.innerHeight / 2);
+        if (!el) return false;
+        let current = el;
+        while (current && current !== document.body) {
+          if (getComputedStyle(current).position === 'fixed') return true;
+          current = current.parentElement;
+        }
+        return false;
+      })()`).catch(() => false);
+      if (hasOverlay) {
+        await page.keyboard.press('Escape').catch(() => {});
+      }
+    }
+    if (autoCleanup.forms) {
+      await page.evaluate(`(() => {
+        document.querySelectorAll('input:not([type=hidden]):not([type=submit])').forEach(el => { el.value = ''; });
+        document.querySelectorAll('textarea').forEach(el => { el.value = ''; });
+        document.querySelectorAll('select').forEach(el => { el.selectedIndex = 0; });
+      })()`).catch(() => {});
+    }
+    if (autoCleanup.scroll) {
+      await page.evaluate('window.scrollTo(0, 0)').catch(() => {});
+    }
+    if (autoCleanup.storage) {
+      await page.evaluate('sessionStorage.clear()').catch(() => {});
+    }
+  }, { auto: true }],
+
+  // Auto-screenshot fixture: captures screenshot on test failure
+  _autoScreenshot: [async ({ page, screenshotOnFailure }, use, testInfo) => {
+    await use();
+
+    if (screenshotOnFailure && testInfo.status !== testInfo.expectedStatus) {
+      await testInfo.attach('screenshot', {
+        body: await page.screenshot(),
+        contentType: 'image/png',
+      });
+    }
+  }, { auto: true }],
 });
 
 export { test, expect };

package/src/orchestrator/templates/specs-test-data.ts

@@ -0,0 +1,66 @@
+import * as fs from 'fs';
+import * as path from 'path';
+import yaml from 'yaml';
+
+export class TestDataLoader {
+  private data: Record<string, any>;
+
+  private constructor(data: Record<string, any>) {
+    this.data = data;
+  }
+
+  /**
+   * Load test data for a screen/feature combination.
+   *
+   * Priority (later wins):
+   * 1. {feature}.yaml — base data
+   * 2. {feature}.{SUNGEN_ENV}.yaml — environment-specific (if SUNGEN_ENV set)
+   */
+  static load(screenName: string, featureName: string): TestDataLoader {
+    const baseDir = path.join(process.cwd(), 'qa', 'screens', screenName, 'test-data');
+    const env = process.env.SUNGEN_ENV;
+
+    let data = loadYamlSync(path.join(baseDir, `${featureName}.yaml`)) || {};
+
+    if (env) {
+      const envData = loadYamlSync(path.join(baseDir, `${featureName}.${env}.yaml`));
+      if (envData) data = deepMerge(data, envData);
+    }
+
+    return new TestDataLoader(data);
+  }
+
+  get(key: string): string {
+    const parts = key.split('.');
+    let current: any = this.data;
+    for (const part of parts) {
+      if (current == null || typeof current !== 'object') {
+        throw new Error(`Test data key not found: ${key} (failed at '${part}')`);
+      }
+      current = current[part];
+    }
+    if (current === undefined || current === null) {
+      throw new Error(`Test data key not found: ${key}`);
+    }
+    return String(current);
+  }
+}
+
+function loadYamlSync(filePath: string): Record<string, any> | null {
+  if (!fs.existsSync(filePath)) return null;
+  const content = fs.readFileSync(filePath, 'utf-8');
+  return yaml.parse(content) || null;
+}
+
+function deepMerge(base: Record<string, any>, override: Record<string, any>): Record<string, any> {
+  const result = { ...base };
+  for (const [key, value] of Object.entries(override)) {
+    if (value && typeof value === 'object' && !Array.isArray(value) &&
+        result[key] && typeof result[key] === 'object' && !Array.isArray(result[key])) {
+      result[key] = deepMerge(result[key], value);
+    } else {
+      result[key] = value;
+    }
+  }
+  return result;
+}