@massu/core 0.1.2 → 0.4.0

package/commands/massu-loop-playwright.md

@@ -0,0 +1,837 @@
+---
+name: massu-loop-playwright
+description: Browser-based audit and fix loop — functional, visual, and performance testing with Playwright
+allowed-tools: Bash(*), Read(*), Write(*), Edit(*), Grep(*), Glob(*), mcp__plugin_playwright_playwright__*, mcp__playwright__*
+---
+name: massu-loop-playwright
+
+# Massu Loop Playwright: Browser-Based Audit & Fix Protocol
+
+**Shared rules**: Read `.claude/commands/_shared-preamble.md` for POST-COMPACTION (CR-35), QUALITY STANDARDS, and CR-9 (fix all issues) rules.
+
+---
+
+## Workflow Position
+
+```
+/massu-loop -> /massu-loop-playwright -> [/massu-simplify] -> /massu-commit -> /massu-push
+(IMPLEMENT)   (BROWSER VERIFY+FIX)      (QUALITY)            (COMMIT)        (PUSH)
+```
+
+**This command runs after implementation (standalone or post-massu-loop). It is the browser-level verification and fix loop.**
+
+**Golden Path Integration**: This command is integrated into `/massu-golden-path` as Phase 2G. When running the golden path, this phase auto-triggers if the plan touches UI files. It can also be run standalone.
+
+---
+
+## USAGE
+
+```
+# Test a specific URL
+/massu-loop-playwright https://example.com/docs
+
+# Test a specific page path (requires a full URL)
+/massu-loop-playwright https://localhost:3000/docs/getting-started
+
+# Test multiple pages
+/massu-loop-playwright https://example.com/docs https://example.com/pricing
+
+# Auto-triggered after massu-loop when plan touches UI files
+(integrated -- no manual invocation needed)
+```
+
+### Argument Parsing
+
+```
+INPUT = $ARGUMENTS
+
+IF INPUT starts with "http":
+  MODE = "single" (or "multi" if multiple URLs provided)
+  TARGET_URL(s) = INPUT (space-separated URLs)
+
+ELSE IF INPUT starts with "/":
+  MODE = "single"
+  WARN: "No base URL configured. Please provide a full URL (e.g., https://example.com{INPUT})"
+  STOP
+
+ELSE:
+  MODE = "single"
+  WARN: "Please provide a full URL (starting with http:// or https://)"
+  STOP
+```
+
+**Note**: Massu is a CLI/package, not a fixed web app. The user MUST provide the full target URL(s) to test. These could be:
+- A documentation site
+- A demo/marketing page
+- Output artifacts generated by Massu (static sites, reports, etc.)
+- A local dev server (`http://localhost:...`)
+
+Alternatively, use `--url` to test specific pages when the target app routes are known.
+
+---
+
+## SAFETY RULE (ABSOLUTE)
+
+**NEVER use real client data for testing. NEVER modify production data during testing.**
+
+- ALL test interactions MUST be read-only or use dedicated test data
+- NEVER submit forms, create records, or trigger actions on production systems
+- NEVER click destructive actions (Delete, Remove, etc.) on live data
+- NEVER create test proposals/orders/emails targeting real clients or contacts
+- NEVER use real client names in test data
+- When a test requires data entry, use clearly fake/test values only
+- If testing a system you do not own, limit to observation and navigation
+
+**Violation of this rule is a P0 incident.**
+
+---
+
+## TOOL SELECTION
+
+Use the Playwright MCP plugin tools (`mcp__plugin_playwright_playwright__*`) as the primary browser automation tool. These provide:
+
+| Tool | Purpose |
+|------|---------|
+| `browser_navigate` | Navigate to URLs |
+| `browser_snapshot` | Capture DOM state for inspection |
+| `browser_take_screenshot` | Capture visual screenshot for evidence |
+| `browser_console_messages` | Retrieve console errors/warnings |
+| `browser_network_requests` | Check for failed network requests |
+| `browser_click` | Click interactive elements |
+| `browser_fill_form` | Fill form fields |
+| `browser_select_option` | Select dropdown options |
+| `browser_press_key` | Keyboard interactions |
+| `browser_hover` | Hover over elements |
+| `browser_tabs` | Manage browser tabs |
+| `browser_wait_for` | Wait for elements/conditions |
+| `browser_evaluate` | Execute JavaScript in browser context |
+| `browser_resize` | Test responsive layouts |
+
+If the plugin tools are unavailable, fall back to standalone `mcp__playwright__*` tools (identical API).
+
+---
+
+## PHASE 0: SETUP
+
+### 0.1 Initialize Report
+
+```
+TIMESTAMP = current time in PST (America/Los_Angeles), format: YYYY-MM-DD-HH-MM
+REPORT_PATH = .claude/playwright-reports/{TIMESTAMP}-{PAGE_SLUG}.md
+```
+
+Create the report directory if needed:
+
+```bash
+mkdir -p .claude/playwright-reports
+```
+
+### 0.2 Navigate to Target URL
+
+```
+Use browser_navigate to: {TARGET_URL}
+```
+
+If the URL includes a path, navigate directly to that path.
+
+### 0.3 Check Application Status
+
+After navigation, use `browser_snapshot` to capture the page state.
+
+| Indicator | Meaning | Action |
+|-----------|---------|--------|
+| Page content visible | App/site loaded | PROCEED |
+| URL redirected to `/login` or auth page | Not logged in | Attempt persistent session or STOP |
+| Error page, blank page, or connection refused | App not running or URL invalid | STOP and report |
+| URL is target path with page content visible | Loaded successfully | PROCEED |
+
+### 0.4 If NOT Ready
+
+If the page requires authentication and persistent session cookies are not available, STOP and output:
+
+```
+AUTHENTICATION REQUIRED
+
+The Playwright browser cannot access the target application.
+
+Please:
+1. The Playwright browser window should be open
+2. Log in to the target application manually in that browser
+3. Wait until you see the expected page loaded
+4. Re-run: /massu-loop-playwright {ORIGINAL_ARGUMENTS}
+
+The command cannot proceed without access to the target page.
+```
+
+**Do NOT attempt to type credentials. Do NOT hardcode passwords. Do NOT proceed without a loaded page.**
+
+### 0.5 If Ready
+
+Report to the user:
+
+```
+Page confirmed loaded. Starting browser audit...
+Mode: {single / multi}
+Target: {URL(s)}
+Pages to test: {N}
+```
+
+---
+
+## PHASE 1: LOAD AUDIT
+
+For each target URL, execute these steps. Capture ALL findings before moving to Phase 2.
+
+### 1.1 Navigate and Wait
+
+```
+1. browser_navigate to the target URL
+2. browser_wait_for page load (up to 15 seconds)
+3. If page does not load in 15 seconds, record as TIMEOUT
+```
+
+### 1.2 Console Error Capture
+
+```
+1. browser_console_messages -- capture ALL messages
+2. Categorize each message:
+```
+
+| Category | Pattern | Severity |
+|----------|---------|----------|
+| **Console Error** | `level: error` | HIGH |
+| **Console Warning** | `level: warning` | MEDIUM |
+| **React Error** | `Uncaught Error`, `Cannot read properties`, hydration mismatch | CRITICAL |
+| **CSP Violation** | `Content Security Policy`, `Refused to`, `blocked by CSP` | HIGH |
+| **Auth Warning** | `401`, `403`, `Unauthorized`, `token expired` | HIGH |
+| **i18n Missing** | `Missing message`, `MISSING_MESSAGE`, `IntlError` | LOW |
+| **Deprecation** | `deprecated`, `will be removed` | LOW |
+
+### 1.3 Network Failure Check
+
+```
+1. browser_network_requests -- check for failed requests
+2. Record any request with status >= 400 or status = 0 (failed to connect)
+```
+
+| Network Issue | Pattern | Severity |
+|---------------|---------|----------|
+| **500 Error** | Status 500+ | CRITICAL |
+| **404 Not Found** | Status 404 on API/resource | HIGH |
+| **CORS Failure** | Status 0, CORS error in console | HIGH |
+| **Timeout** | Request hung > 10s | MEDIUM |
+| **CSP Block** | Resource blocked by CSP | MEDIUM |
+
+### 1.4 CSP Violation Detection
+
+Use `browser_evaluate` to check for CSP violations:
+
+```javascript
+// Check for CSP violation reports
+window.__cspViolations = window.__cspViolations || [];
+document.addEventListener('securitypolicyviolation', (e) => {
+  window.__cspViolations.push({
+    blockedURI: e.blockedURI,
+    violatedDirective: e.violatedDirective,
+    originalPolicy: e.originalPolicy
+  });
+});
+```
+
+After a brief wait, collect violations:
+
+```javascript
+JSON.stringify(window.__cspViolations || []);
+```
+
+### 1.5 Record Load Audit Results
+
+```markdown
+### Load Audit: {URL}
+
+| Check | Result | Details |
+|-------|--------|---------|
+| Console Errors | {N} | [list] |
+| Console Warnings | {N} | [list] |
+| Network Failures | {N} | [list] |
+| CSP Violations | {N} | [list] |
+| Load Time | fast/slow/timeout | {ms if available} |
+```
+
+---
+
+## PHASE 2: INTERACTIVE TESTING
+
+### 2.1 Element Discovery
+
+Use `browser_snapshot` to capture the full DOM state and identify ALL interactive elements:
+
+- **Buttons**: All `<button>` elements and elements with `role="button"`
+- **Links**: All `<a>` elements with `href` attributes
+- **Forms**: All `<form>` elements with their inputs
+- **Selects/Dropdowns**: All `<select>` and custom dropdown components
+- **Tabs**: All tab navigation elements
+- **Modals/Dialogs**: Any trigger elements for sheets, dialogs, modals
+- **Data Tables**: Sortable headers, pagination, filters, row actions
+
+Create an inventory:
+
+```markdown
+### Interactive Element Inventory: {URL}
+
+| # | Element | Type | Text/Label | Testable |
+|---|---------|------|------------|----------|
+| 1 | Button | button | "Get Started" | YES |
+| 2 | Link | a | "Documentation" | YES |
+| 3 | Select | select | "Version" | YES |
+| 4 | Tab | tab | "Overview" | YES |
+```
+
+### 2.2 Systematic Interaction Testing
+
+For EACH testable interactive element:
+
+```
+1. Capture console state BEFORE interaction (browser_console_messages)
+2. Perform the interaction:
+   - Buttons: browser_click
+   - Links: browser_click (check navigation or action)
+   - Selects: browser_select_option (try each option)
+   - Tabs: browser_click (switch between tabs)
+   - Forms: browser_fill_form (with safe test data ONLY)
+3. Wait briefly for async operations (2-3 seconds)
+4. Capture console state AFTER interaction (browser_console_messages)
+5. Compare before/after -- record any NEW errors
+6. browser_snapshot to verify DOM state after interaction
+7. If interaction opened a modal/sheet: test elements inside, then close it
+8. Navigate BACK if the interaction changed the URL (unless testing navigation)
+```
+
+**SAFETY RULES for interactive testing:**
+
+| Rule | Why |
+|------|-----|
+| NEVER submit forms with real data | Could create real records |
+| NEVER click "Delete" or destructive actions | Could destroy production data |
+| NEVER click "Send" on emails/notifications | Would send to real users |
+| NEVER click "Submit" on payment/order forms | Would create real transactions |
+| For forms: fill but do NOT submit | Test validation without side effects |
+| For destructive buttons: test existence, do NOT click | Verify they render, nothing more |
+
+### 2.3 Record Interactive Test Results
+
+For each element tested:
+
+```markdown
+| # | Element | Action | Console Errors After | Visual Issue | Status |
+|---|---------|--------|---------------------|--------------|--------|
+| 1 | "Get Started" button | click | 0 | None | PASS |
+| 2 | "Version" select | select "v2" | 1: TypeError | None | FAIL |
+| 3 | "Overview" tab | click | 0 | Content missing | FAIL |
+```
+
+---
+
+## PHASE 3: VISUAL AUDIT
+
+### 3.1 Broken Image Detection
+
+Use `browser_evaluate` to check for broken images:
+
+```javascript
+Array.from(document.querySelectorAll('img')).filter(img =>
+  !img.complete || img.naturalWidth === 0
+).map(img => ({ src: img.src, alt: img.alt }));
+```
+
+### 3.2 Layout Issue Detection
+
+Use `browser_snapshot` and `browser_take_screenshot` to check for:
+
+| Issue | How to Detect |
+|-------|---------------|
+| **Overflow/Scroll** | Elements extending beyond viewport |
+| **Overlapping** | Z-index issues, elements stacking incorrectly |
+| **Missing Content** | Empty sections that should have content |
+| **Broken Alignment** | Elements not aligned with grid/layout |
+| **Text Truncation** | Important text cut off without ellipsis |
+| **Icon Issues** | Missing or broken icons |
+
+### 3.3 Responsive Check
+
+Use `browser_resize` to test at key breakpoints:
+
+| Breakpoint | Width | Height | What to Check |
+|------------|-------|--------|---------------|
+| Desktop | 1440 | 900 | Full layout |
+| Tablet | 768 | 1024 | Responsive collapse |
+| Mobile | 375 | 812 | Mobile layout |
+
+For each breakpoint:
+1. `browser_resize` to the dimensions
+2. `browser_take_screenshot` for visual evidence
+3. Check for layout breaks, overflow, overlapping elements
+4. Resize back to desktop (1440x900) when done
+
+### 3.4 Screenshot Evidence
+
+Take screenshots at these key states:
+
+| State | When | Save As |
+|-------|------|---------|
+| **Initial Load** | After page fully loads | `{report-dir}/01-initial-load.png` |
+| **After Interaction** | After any interaction reveals an issue | `{report-dir}/02-interaction-{element}.png` |
+| **Responsive** | At each breakpoint if issues found | `{report-dir}/03-responsive-{breakpoint}.png` |
+| **Error State** | When a visual error is visible | `{report-dir}/04-error-{description}.png` |
+
+### 3.5 Record Visual Audit Results
+
+```markdown
+### Visual Audit: {URL}
+
+| # | Issue | Type | Location | Severity | Screenshot |
+|---|-------|------|----------|----------|------------|
+| 1 | Broken image | Image | Hero section | MEDIUM | 04-error-broken-img.png |
+| 2 | Layout overflow | Overflow | Sidebar on mobile | HIGH | 03-responsive-375.png |
+```
+
+---
+
+## PHASE 4: PERFORMANCE AUDIT
+
+### 4.1 Page Load Timing
+
+Use `browser_evaluate` to collect performance metrics:
+
+```javascript
+const perf = performance.getEntriesByType('navigation')[0];
+JSON.stringify({
+  domContentLoaded: Math.round(perf.domContentLoadedEventEnd - perf.startTime),
+  loadComplete: Math.round(perf.loadEventEnd - perf.startTime),
+  ttfb: Math.round(perf.responseStart - perf.requestStart),
+  domInteractive: Math.round(perf.domInteractive - perf.startTime)
+});
+```
+
+### 4.2 Resource Analysis
+
+Use `browser_evaluate` to find large resources:
+
+```javascript
+performance.getEntriesByType('resource')
+  .filter(r => r.transferSize > 500000)
+  .map(r => ({
+    name: r.name.split('/').pop(),
+    url: r.name,
+    size: Math.round(r.transferSize / 1024) + 'KB',
+    duration: Math.round(r.duration) + 'ms',
+    type: r.initiatorType
+  }));
+```
+
+### 4.3 Network Waterfall Check
+
+Use `browser_network_requests` to identify:
+
+- Requests taking > 3 seconds
+- Sequential requests that could be parallelized
+- Duplicate requests to the same endpoint
+- Large payloads (> 1MB)
+
+### 4.4 Performance Thresholds
+
+| Metric | Good | Needs Work | Critical |
+|--------|------|------------|----------|
+| DOM Content Loaded | < 2s | 2-5s | > 5s |
+| Full Load | < 4s | 4-8s | > 8s |
+| TTFB | < 500ms | 500ms-1.5s | > 1.5s |
+| Largest Resource | < 500KB | 500KB-2MB | > 2MB |
+| API Response | < 1s | 1-3s | > 3s |
+
+### 4.5 Record Performance Results
+
+```markdown
+### Performance Audit: {URL}
+
+| Metric | Value | Rating |
+|--------|-------|--------|
+| DOM Content Loaded | {N}ms | GOOD/WARN/CRITICAL |
+| Full Load | {N}ms | GOOD/WARN/CRITICAL |
+| TTFB | {N}ms | GOOD/WARN/CRITICAL |
+| Largest Resource | {name} ({size}) | GOOD/WARN/CRITICAL |
+| Slow API Calls | {N} | GOOD/WARN/CRITICAL |
+```
+
+---
+
+## PHASE 5: FIX LOOP
+
+**This is what differentiates `/massu-loop-playwright` from `/massu-verify-playwright`. This command FIXES issues.**
+
+### 5.1 Triage Findings
+
+Compile ALL issues from Phases 1-4 into a single prioritized list:
+
+| Priority | Criteria | Action |
+|----------|----------|--------|
+| **P0 - CRITICAL** | Console errors, React crashes, 500s, data exposure | Fix immediately in this loop |
+| **P1 - HIGH** | Network failures, CSP violations, broken interactions, auth warnings | Fix in this loop |
+| **P2 - MEDIUM** | Visual issues, performance warnings, broken images | Fix in this loop if code-fixable |
+| **P3 - LOW** | Console warnings, deprecations, i18n missing keys | Document but do not fix unless trivial |
+
+### 5.2 Fix Loop Execution
+
+```
+issues = ALL findings from Phases 1-4, sorted by priority (P0 first)
+fix_count = 0
+
+FOR EACH issue in issues WHERE priority <= P2:
+  1. IDENTIFY the root cause
+     - Use Grep/Glob/Read to find the source file
+     - Trace from browser error to source code
+     - Check CLAUDE.md patterns for known fixes
+
+  2. APPLY the fix
+     - Edit the source file(s)
+     - Follow ALL CLAUDE.md patterns (no shortcuts)
+
+  3. VERIFY the fix
+     - The live site requires a deployment/rebuild to verify browser-side
+     - For code-level verification: run applicable VR-* checks
+     - VR-GREP: Confirm correct pattern is present
+     - VR-NEGATIVE: Confirm incorrect pattern is removed
+     - VR-BUILD: npm run build (must pass)
+     - VR-TYPE: npx tsc --noEmit (must pass)
+
+  4. LOG the fix
+     - Record in fix log table (see report format below)
+     - fix_count += 1
+
+END FOR
+```
+
+### 5.3 Post-Fix Browser Retest
+
+**After ALL code fixes are applied and build/type checks pass:**
+
+```
+1. Reload the target URL in the browser (browser_navigate)
+2. Re-run Phase 1 (Load Audit) on all target URLs
+3. Re-run Phase 2 (Interactive Testing) for elements that had failures
+4. Compare before/after error counts
+
+IF new_errors > 0:
+  Add new errors to issues list
+  Return to 5.2 (fix loop)
+ELSE:
+  PROCEED to Phase 6
+```
+
+**NOTE**: Browser retesting after code fixes requires a rebuild/deployment. If the site is built locally, rebuild and re-check. If deployed via CI/CD, note that a deployment is needed and document the expected fix.
+
+### 5.4 Zero-Issue Standard
+
+The fix loop continues until:
+
+- ALL P0 and P1 issues have code fixes applied
+- ALL P2 issues are either fixed or documented with justification
+- Build passes (exit 0)
+- Type check passes (0 errors)
+- No new console errors introduced by fixes
+
+**Partial fixes are NOT acceptable. Every fixable issue must be addressed.**
+
+---
+
+## PHASE 6: REPORT
+
+### 6.1 Generate Report
+
+Save the complete report to: `.claude/playwright-reports/{TIMESTAMP}-{PAGE_SLUG}.md`
+
+Timestamps MUST be in PST (America/Los_Angeles).
+
+### 6.2 Report Format
+
+```markdown
+# Playwright Audit Report
+
+**URL**: {tested URL(s)}
+**Date**: {YYYY-MM-DD HH:MM PST}
+**Mode**: {single / multi}
+**Status**: PASS / FAIL ({N} issues remaining)
+
+---
+
+## Summary
+
+| Metric | Count |
+|--------|-------|
+| Pages Tested | {N} |
+| Console Errors Found | {N} |
+| Console Warnings Found | {N} |
+| Network Failures Found | {N} |
+| CSP Violations Found | {N} |
+| Broken Images Found | {N} |
+| Interactive Element Failures | {N} |
+| Visual Issues Found | {N} |
+| Performance Issues Found | {N} |
+| **Total Issues** | **{N}** |
+| **Issues Fixed** | **{N}** |
+| **Issues Remaining** | **{N}** |
+
+---
+
+## Console Errors ({N})
+
+| # | Page | Message | Severity | Fixed |
+|---|------|---------|----------|-------|
+| 1 | /docs | TypeError: Cannot read properties of null | CRITICAL | YES |
+
+## Console Warnings ({N})
+
+| # | Page | Message | Severity | Fixed |
+|---|------|---------|----------|-------|
+
+## Network Failures ({N})
+
+| # | Page | URL | Status | Type | Fixed |
+|---|------|-----|--------|------|-------|
+
+## CSP Violations ({N})
+
+| # | Page | Blocked URI | Directive | Fixed |
+|---|------|-------------|-----------|-------|
+
+## Broken Images ({N})
+
+| # | Page | Image Src | Alt Text | Fixed |
+|---|------|-----------|----------|-------|
+
+## Interactive Element Failures ({N})
+
+| # | Page | Element | Action | Error | Fixed |
+|---|------|---------|--------|-------|-------|
+
+## Visual Issues ({N})
+
+| # | Page | Issue | Location | Breakpoint | Fixed |
+|---|------|-------|----------|------------|-------|
+
+## Performance Issues ({N})
+
+| # | Page | Metric | Value | Threshold | Fixed |
+|---|------|--------|-------|-----------|-------|
+
+---
+
+## Fix Log
+
+| # | Issue | Severity | File(s) Changed | Fix Applied | VR Check | Verified |
+|---|-------|----------|-----------------|-------------|----------|----------|
+| 1 | TypeError in DocsPage | P0 | packages/core/src/... | Added null guard | VR-GREP | YES |
+
+---
+
+## Screenshots
+
+| # | Description | Path |
+|---|-------------|------|
+| 1 | Initial page load | .claude/playwright-reports/{dir}/01-initial-load.png |
+
+---
+
+## Unfixed Issues (P3 / Deferred)
+
+| # | Issue | Severity | Reason Not Fixed |
+|---|-------|----------|------------------|
+| 1 | Deprecation warning: findDOMNode | P3 | Third-party library, not actionable |
+
+---
+
+## Post-Fix Verification
+
+| Check | Command | Result |
+|-------|---------|--------|
+| Build | npm run build | Exit 0 / PENDING |
+| Types | npx tsc --noEmit | 0 errors / PENDING |
+| Browser Retest | Console errors after reload | 0 / PENDING DEPLOY |
+```
+
+### 6.3 Present to User
+
+After saving the report, present to the user:
+
+1. Summary (total issues found vs fixed)
+2. Any remaining unfixed issues with justification
+3. The report file path
+4. Whether a rebuild/deployment is needed to verify browser-side fixes
+5. Recommendation for next steps
+
+---
+
+## INTEGRATION WITH MASSU-LOOP
+
+### Auto-Trigger Criteria
+
+When running as part of `/massu-loop`, this command auto-triggers if the plan touches ANY of these file patterns:
+
+```
+packages/*/src/**/*.tsx
+packages/*/src/**/*.ts
+src/**/*.tsx
+src/**/*.ts
+src/**/*.css
+*.html
+```
+
+### Integration Protocol
+
+```
+IF massu-loop plan touches UI files:
+  1. Identify which pages are affected by the changed files
+     - Use file path to determine which output/page is affected
+     - Include parent layout/template pages if layout files changed
+  2. Run /massu-loop-playwright for each affected page
+  3. Include playwright report in massu-loop completion output
+  4. Any P0/P1 issues found become GAPS in the massu-loop audit
+```
+
+---
+
+## COMMON ISSUE PATTERNS
+
+### Issues This Command Catches That Code Review Misses
+
+| Issue | Example | Root Cause |
+|-------|---------|------------|
+| Hydration mismatch | `Text content does not match server-rendered HTML` | Date formatting, random IDs |
+| Missing null guard | `Cannot read properties of undefined (reading 'map')` | Data not loaded yet |
+| CSP violation | `Refused to load script from 'cdn.example.com'` | Missing CSP directive |
+| Broken API call | `Failed to fetch`, `NetworkError` | Endpoint renamed/removed |
+| Layout shift | Content jumps after data loads | Missing skeleton/loading state |
+| Dark mode issue | Element invisible in dark mode | Hardcoded color instead of semantic class |
+| Mobile overflow | Horizontal scroll on mobile | Fixed-width element in flex container |
+| Stale cache | Old data shown after mutation | Missing query invalidation |
+
+### Common Error Patterns
+
+| Error | Likely Fix |
+|-------|-----------|
+| `Hydration failed` | Check for browser-only APIs in server components |
+| `Cannot update during render` | Move state update to useEffect |
+| `Maximum update depth exceeded` | Check useEffect/useCallback dependency arrays |
+| `Each child should have a unique key` | Add stable key prop to list items |
+| `Invalid hook call` | Check component is not called as regular function |
+| `Module not found` | Check import paths, missing dependency |
+
+---
+
+## NON-NEGOTIABLE RULES
+
+1. **NEVER use real client data** -- all test data uses fake/test values only
+2. **NEVER click destructive actions** -- no Delete, Send, Submit on live data
+3. **NEVER skip a phase** -- all 6 phases must execute
+4. **NEVER claim "PASS" with unfixed P0/P1 issues** -- zero-issue standard for critical/high
+5. **Evidence over claims** -- every finding has a console message, screenshot, or DOM snapshot
+6. **Fix in code, verify in browser** -- fixes are applied to source files, not browser hacks
+7. **PST timestamps** -- all report timestamps in America/Los_Angeles timezone
+8. **Complete reports** -- even if zero issues found, generate the full report structure
+9. **Follow CLAUDE.md patterns** -- all code fixes must comply with established patterns
+10. **CR-9 always** -- fix ALL issues encountered, whether from current changes or pre-existing
+
+---
+
+## ERROR HANDLING
+
+### Browser Session Issues
+
+| Scenario | Action |
+|----------|--------|
+| Page returns 404 | Record as issue, check if route exists in codebase |
+| Page hangs (>15s) | Record as TIMEOUT, check for infinite loops |
+| Redirected to login/auth | Session expired. STOP. Report partial results. |
+| Playwright not available | STOP. Tell user to check MCP server. |
+| Console messages unavailable | Use browser_evaluate as fallback to check window.console |
+
+### Fix Loop Safety
+
+| Scenario | Action |
+|----------|--------|
+| Fix introduces new error | Revert fix, try alternative approach |
+| Build fails after fix | Revert fix, investigate build error separately |
+| Fix requires DB change | Document but do NOT apply -- DB changes need separate migration |
+| Fix scope expands beyond page | Document for separate task, fix only the targeted issue |
+| More than 20 issues on one page | Fix P0/P1 first, document P2/P3 for batch fix |
+
+### Circuit Breaker
+
+If the fix loop exceeds 5 iterations on the same page without reaching zero issues:
+
+```
+CIRCUIT BREAKER: Fix loop stalled after {N} iterations on {URL}.
+Remaining issues: {list}
+Options:
+  (a) Continue with different approach
+  (b) Save partial report and stop
+  (c) Escalate remaining issues for manual review
+```
+
+Ask the user how to proceed.
+
+---
+
+## AUTO-LEARNING PROTOCOL (MANDATORY after every fix)
+
+After EVERY browser-discovered bug fix:
+
+### Step 1: Ingest into Memory
+Use `mcp__massu__massu_memory_ingest` with:
+- type: "bugfix"
+- description: "[Browser symptom] -> [Code fix]"
+- files: [list of files changed]
+- importance: 5 (if crash/data issue), 3 (if build/type), 2 (if cosmetic)
+
+### Step 2: Record Pattern
+Update `.claude/session-state/CURRENT.md` with:
+```markdown
+## [Category] Browser Bug (discovered [date])
+- SYMPTOM: [what the browser showed]
+- ROOT CAUSE: [what was wrong in code]
+- FIX: [what was changed]
+- File(s): [where]
+```
+
+### Step 3: Add to Pattern Scanner (if grep-able)
+If the incorrect pattern can be detected by grep, add it to `scripts/massu-pattern-scanner.sh`.
+
+### Step 4: Codebase-Wide Search (CR-9)
+```bash
+grep -rn "[bad_pattern]" packages/ src/ --include="*.ts" --include="*.tsx"
+```
+Fix ALL instances, not just the one found on this page.
+
+---
+
+## START NOW
+
+**Step 0: Write AUTHORIZED_COMMAND to session state (CR-35)**
+
+Before any other work, update `session-state/CURRENT.md` to include:
+```
+AUTHORIZED_COMMAND: massu-loop-playwright
+```
+
+**Then execute:**
+
+1. Parse arguments to determine mode (single URL / multi URL)
+2. Launch Playwright and navigate to target
+3. Verify page is loaded (STOP if not accessible)
+4. **PHASE 1**: Load audit -- console errors, network failures, CSP violations
+5. **PHASE 2**: Interactive testing -- discover and test all interactive elements
+6. **PHASE 3**: Visual audit -- broken images, layout issues, responsive check
+7. **PHASE 4**: Performance audit -- load times, large resources, slow APIs
+8. **PHASE 5**: Fix loop -- triage, fix code, verify, repeat until zero P0/P1 issues
+9. **PHASE 6**: Report -- generate comprehensive report, save to filesystem, present to user
+
+**Remember: Find it. Fix it. Prove it. Report it.**