Validation Gate Automation with expo-mcp

Overview

Execute validation gates with expo-mcp autonomous testing (visual verification) and HARD STOP enforcement on failures.

Core principle: ALL tests must pass. Use expo-mcp for visual verification. STOP immediately if any fail.

Announce at start: "I'm using the claude-mobile-validation-gate skill to execute Validation Gate [X]."

When to Use

• Executing Gate 3A (backend functional testing)
• Executing Gate 4A (frontend visual testing with expo-mcp)
• Executing Gates 6A-E (integration testing with expo-mcp)
• Verifying phase completion
• Re-testing after fixes

Gate Overview

Gate	Type	Primary Tool	Pass Criteria
3A	Backend functional	wscat + filesystem	All 14 tests pass
4A	Frontend visual	expo-mcp autonomous	AI verifies all screens
6A	Connection	expo-mcp visual	Green dot shows "Connected"
6B	Message flow	expo-mcp visual	Message appears, streams, completes
6C	Tool execution	expo-mcp visual	Tool card appears, file created
6D	File browser	expo-mcp visual	Navigation works, file opens
6E	Sessions	expo-mcp visual	Create, switch, delete work

Gate 4A: Frontend Visual Testing (expo-mcp)

Prerequisites: Metro with EXPO_UNSTABLE_MCP_SERVER=1, app built and running

Autonomous Testing Workflow:

Step 1: "Take screenshot of Chat screen empty state"
AI analyzes: Purple gradient ✅, Input field ✅, Send button ✅

Step 2: "Find view with testID 'message-input'"
expo-mcp: {found: true, accessible: true, enabled: true} ✅

Step 3: "Tap button with testID 'send-button'"
expo-mcp: Tapped successfully ✅

Step 4: "Take screenshot showing all 5 screens"
AI verifies: All screens render correctly ✅

Step 5: "Test navigation between screens"
expo-mcp: Navigation works ✅

Step 6: AI determines: PASS or FAIL

Execute Gate 4A

// Manual expo-mcp testing workflow (not bash script)
// Use natural language prompts:

"Take screenshot of Chat screen and verify purple gradient background"
"Find view with testID 'settings-button' and verify it exists"
"Tap button with testID 'settings-button'"
"Take screenshot and verify Settings screen loaded"
"Navigate through all 5 screens and verify each renders correctly"

// AI analyzes each screenshot visually
// AI determines overall PASS/FAIL

Gate 3A: Backend Functional Testing

Execute:

mcp__serena__execute_shell_command({
  command: "./scripts/validate-gate-3a.sh",
  cwd: "/Users/nick/Desktop/claude-mobile-expo"
});

Pass Criteria: Exit code 0, all 14 tests pass

Gates 6A-E: Integration Testing (expo-mcp)

All integration gates use expo-mcp for visual verification:

Gate 6A Example:

"Take screenshot of Chat screen connection status"
AI verifies: Green dot visible ✅, "Connected" text ✅

Gate 6B Example:

"Type 'Hello Claude' in message input"
"Tap send button with testID 'send-button'"
"Take screenshot and verify user message appears"
"Take screenshot and verify assistant response streams in"

HARD STOP Enforcement

If ANY test fails:

// 1. STOP immediately
// 2. Read error logs
mcp__serena__read_file({relative_path: "logs/combined.log"});
mcp__serena__read_file({relative_path: "logs/error.log"});

// 3. Invoke debugging
// Use @systematic-debugging

// 4. Fix issues

// 5. Re-run gate

// 6. Do NOT proceed to next phase until PASS

Save Results (MANDATORY)

mcp__serena__write_memory({
  memory_name: "validation-gate-[X]-results-YYYY-MM-DD",
  content: `
# Gate [X] Results
Status: PASSED ✅

[All test results]
[Screenshots if applicable]
[expo-mcp verification details]

Ready for next phase.
`
});

Common Mistakes

Mistake	Reality
"Skip some tests"	WRONG. ALL tests required.
"One failure is OK"	WRONG. ANY failure = HARD STOP.
"Manual testing is better"	WRONG for Gate 4A. expo-mcp autonomous required.
"Proceed anyway"	WRONG. Fix and re-test.

Red Flags

• "This test is obvious" → WRONG. Run it anyway.
• "One failure won't matter" → WRONG. HARD STOP on ANY failure.
• "I'll fix in integration" → WRONG. Fix now, re-validate.

Integration

• Uses:

  @claude-mobile-ios-testing

  for Gate 4A
• Uses:

  @websocket-integration-testing

  for Gate 3A
• Triggers:

  @systematic-debugging

  if failures occur