dev-test-electron

Announce: "I'm using dev-test-electron for Electron app automation via Chrome DevTools Protocol."

## REAL Test Requirements for Electron Apps

A REAL Electron test must replicate what the user does. FAKE tests test something else.

Before writing ANY test, verify from SPEC.md/PLAN.md:

REAL Test Criteria	Your Test Must
User workflow	Replicate exact steps (click → type → see result)
Protocol	Use SAME protocol as production (WebSocket, IPC, etc.)
UI interaction	Interact with ACTUAL UI elements user sees
Verification	Check what USER sees, not internal state

The Electron-Specific Fake Test Trap

Electron apps often use WebSocket/IPC internally. Testing HTTP is a FAKE test.

FAKE Electron Test	Why It's Fake	REAL Test
HTTP endpoint test	App uses WebSocket	Test WebSocket connection
Direct function call	User clicks button	CDP Input.dispatchMouseEvent or Runtime.evaluate click
Check internal state	User sees panel/status	CDP screenshot or DOM query
Mock IPC layer	Production uses real IPC	Test actual IPC messages
Skip main process	Main process has logic	Test BOTH renderer AND main

Before You Write a Test, Ask:

1. What protocol does this feature use? (WebSocket? IPC? HTTP?)
2. What does the user actually click/type?
3. What does the user actually SEE?
4. Am I testing the SAME code path as production?

If any answer is "I don't know" → Go back to SPEC.md. Don't guess.

Electron Testing Facts

• Production code paths are the test target: users click buttons (CDP Input events, not direct function calls), the app talks over its real transport (WebSocket, not an easier HTTP stand-in), and the user SEES the panel even if its state is "internal". A test that bypasses any of these exercises a code path production never runs — its pass is an unverified claim about the real one.
• The main process and IPC need coverage as much as the renderer: main-process crashes take down the whole app, and IPC bugs fail silently. Renderer-only testing asserts whole-app health from partial evidence.
• Mocking CDP hides exactly the bugs this skill exists to catch; use the real CDP connection.

## Gate Reminder

Before taking screenshots or running E2E tests, you MUST complete all 6 gates from dev-tdd:

GATE 1: BUILD
GATE 2: LAUNCH (with file-based logging)
GATE 3: WAIT
GATE 4: CHECK PROCESS
GATE 5: READ LOGS ← MANDATORY, CANNOT SKIP
GATE 6: VERIFY LOGS
THEN: E2E tests/screenshots

You loaded dev-tdd earlier. Follow the gates now.

Contents

• Tool Availability Gate
• When to Use Electron CDP
• Connecting to Electron
• CDP Domains
• Renderer Process Automation
• Main Process Control
• Verification
• Complete E2E Examples

Electron E2E Testing via Chrome DevTools Protocol

## Tool Availability Gate

Verify CDP tooling is available before proceeding.

Check for these tools:

# Check for curl (CDP communication)
which curl || echo "MISSING: curl"

# Check for jq (JSON parsing)
which jq || echo "MISSING: jq"

# Check for websocat or wscat (WebSocket CLI)
which websocat || which wscat || echo "MISSING: WebSocket CLI"

If missing tools:

STOP: Cannot proceed with Electron CDP automation.

Missing tools needed for CDP:
- curl (for HTTP requests)
- jq (for JSON parsing)
- websocat or wscat (for WebSocket communication)

Install with:
  # macOS: Install via nix-darwin (see ~/nix/). Do NOT use brew.
  # Linux: sudo apt install curl jq websocat

Reply when installed and I'll continue testing.

This gate is non-negotiable. Missing tools = full stop.

## When to Use Electron CDP

USE Electron CDP when you need:

• Test Electron desktop applications (VS Code, Slack, etc.)
• Debug Electron renderer process (console, DOM, network)
• Automate Electron main process (native menus, dialogs, IPC)
• Multi-window Electron testing
• Electron-specific features (webContents, BrowserWindow)
• File system operations from Electron app

DO NOT use Electron CDP when:

• Testing web applications only (use Chrome MCP or Playwright)
• Testing non-Electron desktop apps (use Hammerspoon for macOS, dev-test-linux for Linux)
• Need headless CI/CD for web apps (use Playwright MCP)

For web apps or native desktop apps, discover and read the relevant skill: Related skills:

• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-chrome/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-playwright/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-hammerspoon/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-linux/SKILL.md and follow its instructions.
• Chrome MCP skill - web debugging
• Playwright skill - headless CI/CD
• Hammerspoon skill - macOS native
• Linux skill - Linux native

Tool Boundary Facts

• Chrome MCP cannot attach to Electron's main process, and Playwright MCP targets web browsers — neither reaches Electron's main process. Electron-specific CDP is the only tool that covers both main process and renderer (see Capability Comparison below); picking a familiar web tool instead produces a test blind to file dialogs, native menus, and main-process crashes.

Capability Comparison

Capability	Electron CDP	Chrome MCP	Playwright MCP	Hammerspoon
Electron renderer	✅	❌	❌	❌
Electron main process	✅	❌	❌	❌
Native menus/dialogs	✅	❌	❌	✅ (macOS only)
Multi-window Electron	✅	❌	❌	✅ (macOS only)
Console/network debugging	✅	✅ (web only)	❌	❌
Headless mode	✅	❌	✅ (web only)	❌
WebSocket IPC	✅	❌	❌	❌

Connecting to Electron

### The Iron Law of Connection

EVERY Electron E2E test MUST establish CDP connection BEFORE any automation.

You CANNOT automate without:

1. Launching Electron with CDP enabled
2. Discovering the CDP WebSocket URL
3. Connecting to the WebSocket
4. Verifying the connection works

Action	Why It Fails Without Connection
Send CDP command	No connection = command never sent
Read console logs	Can't receive events without WebSocket
Navigate to page	CDP Page.navigate requires connection
Take screenshot	CDP Page.captureScreenshot requires connection

"App is running" ≠ "CDP is connected". Verify connection first.

Enable Remote Debugging

Launch Electron with remote debugging port:

# Option 1: Fixed port
/path/to/electron-app --remote-debugging-port=9222

# Option 2: Random port (app outputs port number)
/path/to/electron-app --remote-debugging-port=0

# Option 3: With logging
/path/to/electron-app --remote-debugging-port=9222 --enable-logging --log-file=/tmp/electron.log 2>&1 &

CRITICAL: For GATE 2 (LAUNCH), always use --log-file flag for file-based logging.

Discover CDP WebSocket URL

# Get list of inspectable targets
curl -s http://localhost:9222/json/list | jq '.'

# Extract WebSocket URL for main target
WS_URL=$(curl -s http://localhost:9222/json/list | jq -r '.[0].webSocketDebuggerUrl')
echo "WebSocket URL: $WS_URL"

Example response:

[
  {
    "description": "",
    "devtoolsFrontendUrl": "/devtools/inspector.html?ws=localhost:9222/devtools/page/...",
    "id": "page-id",
    "title": "My Electron App",
    "type": "page",
    "url": "file:///app/index.html",
    "webSocketDebuggerUrl": "ws://localhost:9222/devtools/page/..."
  }
]

Connect via WebSocket

# Interactive WebSocket session
websocat "$WS_URL"

# Send CDP commands (one per line, JSON format)
{"id":1,"method":"Runtime.enable"}
{"id":2,"method":"Page.enable"}
{"id":3,"method":"Runtime.evaluate","params":{"expression":"document.title"}}

Helper script: See scripts/connect-electron-cdp.sh for automated connection.

CDP Domains

Essential Domains for Electron

Domain	Purpose	Example
Runtime	Execute JavaScript, console logs	Runtime.evaluate, Runtime.consoleAPICalled
Page	Navigation, screenshots, DOM events	Page.navigate, Page.captureScreenshot
DOM	Query and manipulate DOM	DOM.getDocument, DOM.querySelector
Debugger	Breakpoints, step debugging	Debugger.setBreakpoint
Network	Network requests/responses	Network.enable, Network.responseReceived
Input	Keyboard/mouse events	Input.dispatchKeyEvent, Input.dispatchMouseEvent

Enable domains before use:

{"id":1,"method":"Runtime.enable"}
{"id":2,"method":"Page.enable"}
{"id":3,"method":"DOM.enable"}
{"id":4,"method":"Network.enable"}

Renderer Process Automation

Execute JavaScript

# Evaluate JavaScript expression
echo '{"id":1,"method":"Runtime.evaluate","params":{"expression":"document.title"}}' | websocat "$WS_URL"

# Execute with return value
echo '{"id":2,"method":"Runtime.evaluate","params":{"expression":"2 + 2","returnByValue":true}}' | websocat "$WS_URL"

# Execute complex script
SCRIPT='document.querySelector("#username").value = "testuser"'
echo "{\"id\":3,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$SCRIPT\"}}" | websocat "$WS_URL"

Navigate to URL

# Navigate to URL (file:// or http://)
echo '{"id":10,"method":"Page.navigate","params":{"url":"file:///app/index.html"}}' | websocat "$WS_URL"

# Wait for load event
echo '{"id":11,"method":"Page.enable"}' | websocat "$WS_URL"
# Listen for Page.loadEventFired event

Read Console Messages

# Enable Runtime domain to receive console events
echo '{"id":20,"method":"Runtime.enable"}' | websocat "$WS_URL"

# Console events arrive as:
# {"method":"Runtime.consoleAPICalled","params":{"type":"log","args":[...]}}

For complete console reading, see references/cdp-api.md

Take Screenshots

# Capture viewport screenshot (PNG base64)
echo '{"id":30,"method":"Page.captureScreenshot"}' | websocat "$WS_URL" > response.json

# Extract base64 and decode
jq -r '.result.data' response.json | base64 -d > screenshot.png

Query DOM

# Get document root
echo '{"id":40,"method":"DOM.getDocument"}' | websocat "$WS_URL"

# Query selector
ROOT_ID=$(jq -r '.result.root.nodeId' response.json)
echo "{\"id\":41,\"method\":\"DOM.querySelector\",\"params\":{\"nodeId\":$ROOT_ID,\"selector\":\"#submit-btn\"}}" | websocat "$WS_URL"

Main Process Control

### Electron Main Process vs Renderer Process

Electron has TWO processes:

Process	What It Does	How to Test
Main	Node.js runtime, native APIs, file system, menus, dialogs	CDP Runtime.evaluate in main context OR IPC
Renderer	Browser/Chromium runtime, web content, DOM	CDP commands (Page, DOM, Runtime)

Both processes MUST be tested. Renderer-only testing is incomplete.

Access Main Process via CDP

Some Electron apps expose main process debugging:

# Check for main process target
curl -s http://localhost:9222/json/list | jq '.[] | select(.type == "node")'

If main process is available:

# Get main process WebSocket URL
MAIN_WS=$(curl -s http://localhost:9222/json/list | jq -r '.[] | select(.type == "node") | .webSocketDebuggerUrl')

# Execute Node.js code in main process
echo '{"id":1,"method":"Runtime.evaluate","params":{"expression":"process.version"}}' | websocat "$MAIN_WS"

Trigger Native Dialogs (via IPC)

# From renderer, send IPC to main process
SCRIPT='require("electron").ipcRenderer.send("open-file-dialog")'
echo "{\"id\":50,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$SCRIPT\"}}" | websocat "$WS_URL"

For advanced main process patterns, see references/electron-specific.md

Verification

### The Iron Law of Verification

EVERY CDP command must be VERIFIED. Sending the command is not enough.

After sending a CDP command, you MUST:

1. Read the response
2. Check for errors (error field in response)
3. Verify the result matches expectations
4. Document the verification

Command	Verification
Runtime.evaluate	Check result.value or result.exceptionDetails
Page.navigate	Wait for Page.loadEventFired event
Page.captureScreenshot	Verify result.data exists and decode base64
DOM.querySelector	Check result.nodeId exists (not 0)

"I sent the command" is not verification. Read the response and verify success.

Response Verification Pattern

# Send command and capture response
RESPONSE=$(echo '{"id":100,"method":"Runtime.evaluate","params":{"expression":"2 + 2"}}' | websocat --one-message "$WS_URL")

# Check for error
if echo "$RESPONSE" | jq -e '.error' > /dev/null; then
    echo "ERROR: CDP command failed"
    echo "$RESPONSE" | jq '.error'
    exit 1
fi

# Verify result
RESULT=$(echo "$RESPONSE" | jq -r '.result.result.value')
if [ "$RESULT" != "4" ]; then
    echo "ERROR: Expected 4, got $RESULT"
    exit 1
fi

echo "✓ VERIFIED: 2 + 2 = $RESULT"

Event-Based Verification

# Enable Page domain
echo '{"id":1,"method":"Page.enable"}' | websocat "$WS_URL" &

# Navigate and wait for load event
echo '{"id":2,"method":"Page.navigate","params":{"url":"file:///app/index.html"}}' | websocat "$WS_URL"

# Wait for Page.loadEventFired event (listen to WebSocket)
# Event format: {"method":"Page.loadEventFired","params":{...}}

Complete E2E Examples

Basic Electron App Test (All 6 Gates)

#!/bin/bash
set -e

# ============ GATE 1: BUILD ============
echo "GATE 1: Building Electron app..."
cd /path/to/electron-app
npm run build
echo "✓ GATE 1 PASSED"

# ============ GATE 2: LAUNCH ============
echo "GATE 2: Launching with CDP and logging..."
npm start -- --remote-debugging-port=9222 --enable-logging --log-file=/tmp/electron.log 2>&1 &
APP_PID=$!
echo "✓ GATE 2 PASSED (PID: $APP_PID)"

# ============ GATE 3: WAIT ============
echo "GATE 3: Waiting for Electron initialization..."
sleep 3
echo "✓ GATE 3 PASSED"

# ============ GATE 4: CHECK PROCESS ============
echo "GATE 4: Checking Electron process..."
if ! ps -p $APP_PID > /dev/null; then
    echo "✗ GATE 4 FAILED: Electron process crashed"
    echo "Reading logs from GATE 5..."
    cat /tmp/electron.log
    exit 1
fi

# Verify CDP port is open
if ! curl -s http://localhost:9222/json/list > /dev/null; then
    echo "✗ GATE 4 FAILED: CDP port not accessible"
    cat /tmp/electron.log
    exit 1
fi
echo "✓ GATE 4 PASSED"

# ============ GATE 5: READ LOGS ============
echo "GATE 5: Reading full runtime logs..."
echo "=== ELECTRON RUNTIME LOGS ==="
cat /tmp/electron.log
echo "=== END LOGS ==="
echo "✓ GATE 5 PASSED (logs read)"

# ============ GATE 6: VERIFY LOGS ============
echo "GATE 6: Verifying no errors in logs..."
if grep -qE "(ERROR|FATAL|CRITICAL|Segmentation|core dumped|Uncaught Exception)" /tmp/electron.log; then
    echo "✗ GATE 6 FAILED: Errors found in logs"
    exit 1
fi
echo "✓ GATE 6 PASSED"

# ============ NOW: E2E TESTING ============
echo "All gates passed. Proceeding to E2E tests..."

# Get WebSocket URL
WS_URL=$(curl -s http://localhost:9222/json/list | jq -r '.[0].webSocketDebuggerUrl')
echo "CDP WebSocket: $WS_URL"

# Enable Runtime domain
echo '{"id":1,"method":"Runtime.enable"}' | websocat --one-message "$WS_URL"

# Execute test: Get document title
RESPONSE=$(echo '{"id":2,"method":"Runtime.evaluate","params":{"expression":"document.title","returnByValue":true}}' | websocat --one-message "$WS_URL")

# Verify response
if echo "$RESPONSE" | jq -e '.error' > /dev/null; then
    echo "✗ E2E FAILED: CDP command error"
    echo "$RESPONSE" | jq '.error'
    exit 1
fi

TITLE=$(echo "$RESPONSE" | jq -r '.result.result.value')
echo "✓ E2E VERIFIED: Document title = '$TITLE'"

# Take screenshot
SCREENSHOT_RESPONSE=$(echo '{"id":3,"method":"Page.captureScreenshot"}' | websocat --one-message "$WS_URL")
echo "$SCREENSHOT_RESPONSE" | jq -r '.result.data' | base64 -d > /tmp/electron_screenshot.png
echo "✓ Screenshot saved: /tmp/electron_screenshot.png"

# Cleanup
kill $APP_PID
echo "✓ E2E TEST PASSED"

Tool description: Execute all 6 gates, then run Electron E2E test with CDP

Form Automation Example

#!/bin/bash
# Assumes gates 1-6 already passed and WS_URL is set

# Enable domains
echo '{"id":1,"method":"Runtime.enable"}' | websocat --one-message "$WS_URL"
echo '{"id":2,"method":"Page.enable"}' | websocat --one-message "$WS_URL"

# Fill form field
FILL_USERNAME='document.querySelector("#username").value = "testuser"'
RESPONSE=$(echo "{\"id\":10,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$FILL_USERNAME\"}}" | websocat --one-message "$WS_URL")

if echo "$RESPONSE" | jq -e '.error' > /dev/null; then
    echo "✗ FAILED: Could not fill username"
    exit 1
fi

FILL_PASSWORD='document.querySelector("#password").value = "testpass"'
echo "{\"id\":11,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$FILL_PASSWORD\"}}" | websocat --one-message "$WS_URL"

# Click submit button
CLICK_SUBMIT='document.querySelector("#submit-btn").click()'
echo "{\"id\":12,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$CLICK_SUBMIT\"}}" | websocat --one-message "$WS_URL"

# Wait for navigation
sleep 1

# Verify success message appears
CHECK_SUCCESS='document.querySelector(".success-message") !== null'
VERIFY_RESPONSE=$(echo "{\"id\":13,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$CHECK_SUCCESS\",\"returnByValue\":true}}" | websocat --one-message "$WS_URL")

SUCCESS=$(echo "$VERIFY_RESPONSE" | jq -r '.result.result.value')
if [ "$SUCCESS" != "true" ]; then
    echo "✗ VERIFICATION FAILED: Success message not found"
    exit 1
fi

echo "✓ VERIFIED: Form submission successful"

# Screenshot for evidence
echo '{"id":14,"method":"Page.captureScreenshot"}' | websocat --one-message "$WS_URL" | jq -r '.result.data' | base64 -d > /tmp/form_success.png
echo "✓ Screenshot: /tmp/form_success.png"

Multi-Window Testing

# Get all inspectable targets
curl -s http://localhost:9222/json/list | jq '.[] | {title: .title, url: .url, wsUrl: .webSocketDebuggerUrl}'

# Connect to specific window by title
WINDOW_WS=$(curl -s http://localhost:9222/json/list | jq -r '.[] | select(.title == "Settings Window") | .webSocketDebuggerUrl')

# Automate the settings window
echo '{"id":1,"method":"Runtime.evaluate","params":{"expression":"document.querySelector(\"#theme\").value = \"dark\""}}' | websocat --one-message "$WINDOW_WS"

For more advanced patterns, see references/advanced-patterns.md

Error Handling

Common CDP Errors

Error	Cause	Solution
Connection refused	Electron not started with --remote-debugging-port	Restart with flag
WebSocket timeout	App crashed or port blocked	Check GATE 4 (process) and GATE 5 (logs)
"error":{"code":-32601}	Method not found	Enable domain first (e.g., Runtime.enable)
exceptionDetails in result	JavaScript error in evaluated code	Check expression syntax
Empty response	WebSocket closed	Reconnect to WebSocket

Retry Pattern

# Retry CDP command up to 3 times
for i in {1..3}; do
    RESPONSE=$(echo "$CDP_COMMAND" | websocat --one-message "$WS_URL")

    if echo "$RESPONSE" | jq -e '.result' > /dev/null; then
        echo "✓ Command succeeded on attempt $i"
        break
    fi

    if [ $i -eq 3 ]; then
        echo "✗ Command failed after 3 attempts"
        echo "$RESPONSE"
        exit 1
    fi

    echo "Retry $i failed, waiting 1s..."
    sleep 1
done

Limitations

### What Electron CDP Cannot Do

Need	Why Electron CDP Fails	Use Instead
Native macOS window management	CDP doesn't control OS	Hammerspoon (macOS)
Cross-platform native automation	CDP is Chromium-only	Platform-specific tools
Test non-Electron apps	CDP requires Electron/Chromium	Hammerspoon, dev-test-linux
Headless CI/CD for web apps	Electron is for desktop apps	Playwright MCP

For web apps, use Playwright or Chrome MCP. For native desktop, use platform tools.

Additional Resources

Reference Files

For detailed CDP API documentation and Electron-specific features:

• references/cdp-api.md - Complete CDP domains reference (Runtime, Page, DOM, Network, Input, Debugger)
• references/electron-specific.md - Electron main process, IPC, native APIs, file dialogs
• references/advanced-patterns.md - Multi-window, devtools, event listeners, WebSocket streaming

Example Files

Working examples in examples/:

• basic-test.sh - Complete E2E test with all 6 gates
• cdp-commands.json - Common CDP command reference

Scripts

Utility scripts in scripts/:

• connect-electron-cdp.sh - Automated CDP connection discovery
• launch-electron-with-logging.sh - Launch template with proper logging
• verify-electron-process.sh - Health check for main + renderer

VS Code Extension Testing (Common Case)

**VS Code extensions are a common Electron test case. Here's how to test them REAL.**

What Makes VS Code Extension Tests REAL

User Action	FAKE Test	REAL Test
Highlight text in editor	editor.setSelection() programmatically	CDP simulate actual text selection
Click Claude panel	Call panel function directly	CDP click on actual panel element
See status in panel	Check internal state variable	CDP query panel DOM for displayed text
Extension uses WebSocket	Test HTTP endpoint	Test WebSocket connection

VS Code Extension Protocol Discovery

Before testing, discover what protocol the extension uses:

# Search for WebSocket usage
rg "WebSocket|ws://" --type ts

# Search for HTTP usage
rg "fetch|axios|http" --type ts

# Search for IPC usage
rg "ipcRenderer|ipcMain" --type ts

If extension uses WebSocket → Your test MUST use WebSocket, not HTTP.

Example: Testing Selection → Panel Status

FAKE test (DON'T DO THIS):

// FAKE: Calls function directly, checks internal state
const selection = await vscode.window.activeTextEditor.selection;
await extensionApi.updateSelection(selection);  // Direct call!
expect(internalState.selectionCount).toBe(5);   // Internal state!

REAL test (DO THIS):

# REAL: Simulates user, checks what user sees

# 1. Use CDP to simulate text selection in editor
SCRIPT='
  const editor = document.querySelector(".monaco-editor");
  // Simulate actual selection via CDP Input events
'
echo "{\"id\":1,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$SCRIPT\"}}" | websocat "$WS_URL"

# 2. Use CDP to query Claude panel for displayed status
VERIFY='document.querySelector(".claude-panel .status-text").textContent'
RESULT=$(echo "{\"id\":2,\"method\":\"Runtime.evaluate\",\"params\":{\"expression\":\"$VERIFY\",\"returnByValue\":true}}" | websocat --one-message "$WS_URL")

# 3. Verify user-visible output
STATUS=$(echo "$RESULT" | jq -r '.result.result.value')
if [[ "$STATUS" != *"5 lines selected"* ]]; then
    echo "✗ FAKE TEST: Panel doesn't show expected status"
    exit 1
fi
echo "✓ REAL TEST: Panel shows '$STATUS'"

VS Code Extension Test Checklist

Before writing VS Code extension test, verify:

[ ] Protocol discovered (WebSocket/HTTP/IPC)
[ ] User workflow documented (what user clicks/sees)
[ ] Test uses SAME protocol as extension
[ ] Test simulates ACTUAL user actions (not API calls)
[ ] Test verifies PANEL DISPLAY (not internal state)
[ ] Test covers BOTH main and renderer processes

If any box is unchecked → Your test is probably FAKE.

Integration

This skill is referenced by dev-test for Electron desktop application testing.

Related skills:

• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-chrome/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-playwright/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-test-hammerspoon/SKILL.md and follow its instructions.
• Read ${CLAUDE_SKILL_DIR}/../../skills/dev-tdd/SKILL.md and follow its instructions.
• Chrome MCP skill - web debugging
• Playwright skill - headless web CI/CD
• Hammerspoon skill - macOS native apps
• TDD skill - TDD protocol and gate enforcement