Agent-Skills.md

Agent Skills: Chrome Debugging and Browser Manipulation via Remote Debugging Protocol

Use when debugging web applications in chrome via the remote debugging protocol. Provides capabilities for inspecting DOM, executing JS, taking screenshots, and automating browser interactions.

UncategorizedID: zenobi-us/dotfiles/chrome-debug

Repository

zenobi-us/dotfiles
zenobi-usLicense: MIT
676

Install this agent skill to your local

pnpm dlx add-skill https://github.com/zenobi-us/dotfiles/tree/HEAD/devtools/files/pi/agent/bundles/developer/skills/browsers/chrome-debug

Skill Files

Browse the full folder contents for chrome-debug.

Download Skill

Loading file tree…

devtools/files/pi/agent/bundles/developer/skills/browsers/chrome-debug/SKILL.md

Skill Metadata

Name
chrome-debug
Description
Use when debugging web applications in chrome via the remote debugging protocol. Provides capabilities for inspecting DOM, executing JS, taking screenshots, and automating browser interactions.

Chrome Debugging and Browser Manipulation via Remote Debugging Protocol

Overview

Chrome DevTools Protocol (CDP) enables remote browser automation and debugging through mcporter.

Key capabilities:

  • Live browser debugging alongside Agent conversations
  • Automated form filling and interaction testing
  • Visual feedback via screenshots
  • Console log and network request inspection
  • JavaScript execution in page context

Prerequisites [CRITICAL]

Before using Chrome DevTools, ensure:

  1. Chrome/Chromium is running with remote debugging enabled
  2. The browser is listening on port 9222 (default)
  3. Test connection with:
mise x node@20 -- mcporter call chrome-devtools.list_pages

If this fails:

  • Start Chrome: google-chrome --remote-debugging-port=9222
  • Check no other process is using port 9222
  • Get a human to help with browser setup

Available Tools

| Tool | Purpose | |------|---------| | list_pages | List all open pages/tabs | | select_page | Select a specific page/tab to work with | | new_page | Create a new browser page/tab | | close_page | Close a browser page/tab | | navigate_page | Navigate to a URL, back, forward, or reload | | take_snapshot | Take a DOM snapshot for inspection (returns UIDs) | | take_screenshot | Capture a screenshot of the current page | | click | Click an element on the page | | fill | Fill input fields with text | | hover | Hover over an element | | press_key | Press keyboard keys (Enter, Tab, Escape, etc.) | | evaluate_script | Execute JavaScript code in the page context | | wait_for | Wait for elements, navigation, or conditions | | list_console_messages | Get all console messages (logs, errors, warnings) | | list_network_requests | Get all network requests made by the page | | emulate | Emulate device settings (network, CPU throttling) | | resize_page | Resize the browser viewport | | performance_start_trace | Start performance tracing | | performance_stop_trace | Stop performance tracing and get results |

[!TIP] Get full tool list: mcporter list chrome-devtools --json | jq -r '.tools[] | [.name, .description] | @tsv' | column -t -s $'\t'

Core Concepts

1. Page Selection Model

  • Chrome DevTools works with multiple pages/tabs
  • Use list_pages to see all open pages
  • Use select_page to choose which page to work with
  • All subsequent commands operate on the selected page

2. UID-Based Element Model [CRITICAL]

  • You CANNOT interact with elements using CSS selectors directly
  • Must first call take_snapshot to get accessibility tree with UIDs
  • UIDs are temporary identifiers for elements (e.g., "5", "12", "42")
  • UIDs are invalidated on navigation - take new snapshot after nav

3. JSON Arguments Required

  • All mcporter commands require --args with JSON object
  • Property names are camelCase (e.g., filePath, fullPage, pageIdx)
  • Never use individual flags like --file-path or --full-page

4. Function-Based Script Evaluation

  • evaluate_script requires a function declaration, not plain code
  • Return values must be JSON-serializable
  • Can pass element arguments via args array with UIDs

Quick Reference

Essential commands in bash-friendly format:

# List and select page
mise x node@20 -- mcporter call chrome-devtools.list_pages
mise x node@20 -- mcporter call chrome-devtools.select_page --args '{"pageIdx":0}'

# Take snapshot to get UIDs
mise x node@20 -- mcporter call chrome-devtools.take_snapshot

# Take screenshot
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"filePath":"./screen.png"}'

# Take full-page screenshot
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"filePath":"./full.png","fullPage":true}'

# Navigate to URL
mise x node@20 -- mcporter call chrome-devtools.navigate_page --args '{"type":"url","url":"http://localhost:3000"}'

# Navigate back/forward/reload
mise x node@20 -- mcporter call chrome-devtools.navigate_page --args '{"type":"back"}'
mise x node@20 -- mcporter call chrome-devtools.navigate_page --args '{"type":"reload"}'

# Click element (requires UID from snapshot)
mise x node@20 -- mcporter call chrome-devtools.click --args '{"uid":"12"}'

# Fill input field
mise x node@20 -- mcporter call chrome-devtools.fill --args '{"uid":"5","value":"test@example.com"}'

# Hover element
mise x node@20 -- mcporter call chrome-devtools.hover --args '{"uid":"8"}'

# Press key
mise x node@20 -- mcporter call chrome-devtools.press_key --args '{"key":"Enter"}'

# Run JavaScript
mise x node@20 -- mcporter call chrome-devtools.evaluate_script --args '{"function":"() => { return document.title }"}'

# Run JS with element argument
mise x node@20 -- mcporter call chrome-devtools.evaluate_script --args '{"function":"(el) => { return el.innerText }","args":[{"uid":"12"}]}'

# List console messages
mise x node@20 -- mcporter call chrome-devtools.list_console_messages

# List only errors
mise x node@20 -- mcporter call chrome-devtools.list_console_messages --args '{"types":["error"]}'

# List network requests
mise x node@20 -- mcporter call chrome-devtools.list_network_requests

# Filter network by type
mise x node@20 -- mcporter call chrome-devtools.list_network_requests --args '{"types":["fetch","xhr"]}'

# Wait for text to appear
mise x node@20 -- mcporter call chrome-devtools.wait_for --args '{"text":"Success"}'

# Emulate network conditions
mise x node@20 -- mcporter call chrome-devtools.emulate --args '{"networkConditions":"Slow 3G"}'

Common Workflows

Basic Element Interaction

# 1. Select page and take snapshot
mise x node@20 -- mcporter call chrome-devtools.list_pages
mise x node@20 -- mcporter call chrome-devtools.select_page --args '{"pageIdx":0}'
SNAPSHOT=$(mise x node@20 -- mcporter call chrome-devtools.take_snapshot)
echo "$SNAPSHOT"

# 2. Find element UID in snapshot output
# Example: uid=12 input type="email"

# 3. Interact with element using its UID
mise x node@20 -- mcporter call chrome-devtools.fill --args '{"uid":"12","value":"user@example.com"}'
mise x node@20 -- mcporter call chrome-devtools.click --args '{"uid":"15"}'

Screenshot Workflow

# Take viewport screenshot
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"filePath":"./screen.png"}'

# Take full-page screenshot
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"filePath":"./full.png","fullPage":true}'

# Screenshot specific element
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"uid":"20","filePath":"./button.png"}'

Debug JavaScript Errors

# Check console for errors
mise x node@20 -- mcporter call chrome-devtools.list_console_messages --args '{"types":["error","warn"]}'

# Check network requests
mise x node@20 -- mcporter call chrome-devtools.list_network_requests --args '{"types":["fetch","xhr"]}'

Run Performance Tests

# Execute JavaScript to get performance metrics
mise x node@20 -- mcporter call chrome-devtools.evaluate_script \
  --args '{"function":"() => { const perf = performance.getEntriesByType(\"navigation\")[0]; return { loadTime: perf.loadEventEnd - perf.fetchStart, domInteractive: perf.domInteractive - perf.fetchStart }; }"}'

Important Reminders

UID Workflow is Mandatory

# ❌ WRONG - CSS selectors don't work
mise x node@20 -- mcporter call chrome-devtools.click --selector "#login-button"

# ✅ CORRECT - Use UIDs from snapshot
mise x node@20 -- mcporter call chrome-devtools.take_snapshot  # Get UIDs first
mise x node@20 -- mcporter call chrome-devtools.click --args '{"uid":"12"}'

UIDs Expire on Navigation

# After navigation, UIDs are invalid
mise x node@20 -- mcporter call chrome-devtools.navigate_page --args '{"type":"url","url":"..."}'

# Take fresh snapshot to get new UIDs
mise x node@20 -- mcporter call chrome-devtools.take_snapshot

Always Use --args with JSON

# ❌ WRONG - Individual flags don't work
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --file-path "./screen.png"

# ✅ CORRECT - Use --args with JSON
mise x node@20 -- mcporter call chrome-devtools.take_screenshot --args '{"filePath":"./screen.png"}'

Quick Troubleshooting

| Error | Solution | |-------|----------| | "Element not found" / "Invalid UID" | Take fresh snapshot: take_snapshot | | "No page selected" | Select page: select_page --args '{"pageIdx":0}' | | "Connection refused" | Start Chrome: google-chrome --remote-debugging-port=9222 | | Screenshot not created | Ensure directory exists and use --args format | | UIDs not working | UIDs expired after navigation - take new snapshot |

Additional Resources

Lazy-load these references based on your needs:

| Reference | When to Use | |-----------|-------------| | Element Interaction | When working with UIDs, clicking, hovering, or measuring elements | | Form Filling | When filling forms, submitting data, or handling keyboard input | | Screenshots | When capturing screenshots, visual testing, or documenting state | | Performance | When measuring page performance, network timing, or emulating conditions | | Debugging | When investigating console errors, network failures, or script evaluation | | Navigation | When navigating pages, managing tabs, or handling viewports | | Troubleshooting | When encountering errors or unexpected behavior |

[!IMPORTANT] Load references only when needed - Don't read all files upfront. Read the specific reference that matches your current task.

Real-World Impact

Integrating Chrome DevTools Protocol enables:

  • Live browser debugging alongside Agent conversations
  • Automated form filling and interaction testing
  • Visual feedback on application behavior
  • Immediate error diagnostics from console logs
  • Screenshot-based validation workflows

Without this integration, debugging web applications requires constant context-switching between browser and Agent.