nixos/shared/linked-dotfiles/opencode/skills/browser-automation

Browser Automation Skill

Control Chrome browser via DevTools Protocol using the use_browser MCP tool.

Structure

browser-automation/
├── SKILL.md                      # Main skill (324 lines, 1050 words)
└── references/
    ├── examples.md               # Complete workflows (672 lines)
    ├── troubleshooting.md        # Error handling (546 lines)
    └── advanced.md               # Advanced patterns (678 lines)

Quick Start

The skill provides:

• Core patterns: Navigate, wait, interact, extract
• Form automation: Multi-step forms, validation, submission
• Data extraction: Tables, structured data, batch operations
• Multi-tab workflows: Cross-site data correlation
• Dynamic content: AJAX waiting, infinite scroll, modals

Installation

This skill requires the use_browser MCP tool from the superpowers-chrome package.

Option 1: Use superpowers-chrome directly

/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers-chrome@superpowers-marketplace

Option 2: Install as standalone skill

Copy this skill directory to your OpenCode skills directory:

cp -r browser-automation ~/.opencode/skills/

Then configure the chrome MCP server in your Claude Desktop config per the superpowers-chrome installation guide.

Usage

The skill is automatically loaded when OpenCode starts. It will be invoked when you:

• Request web automation tasks
• Need to fill forms
• Want to extract content from websites
• Mention Chrome or browser control

Example prompts:

• "Fill out the registration form at example.com"
• "Extract all product names and prices from this page"
• "Navigate to my email and find the receipt from yesterday"

Contents

SKILL.md

Main reference with:

• Quick reference table for all actions
• Core workflow patterns
• Common mistakes and solutions
• Real-world impact metrics

references/examples.md

Complete workflows including:

• E-commerce booking flows
• Multi-step registration forms
• Price comparison across sites
• Data extraction patterns
• Multi-tab operations
• Dynamic content handling
• Authentication workflows

references/troubleshooting.md

Solutions for:

• Element not found errors
• Timeout issues
• Click failures
• Form submission problems
• Tab index errors
• Extract returning empty

Plus best practices for selectors, waiting, and debugging.

references/advanced.md

Advanced techniques:

• Network interception
• JavaScript injection
• Complex waiting patterns
• Data manipulation
• State management
• Visual testing
• Performance monitoring
• Accessibility testing
• Frame handling

Progressive Disclosure

The skill uses progressive disclosure to minimize context usage:

1. SKILL.md loads first - quick reference and common patterns
2. examples.md - loaded when implementing specific workflows
3. troubleshooting.md - loaded when encountering errors
4. advanced.md - loaded for complex requirements

Key Features

Single Tool Interface

All operations use one tool with action-based parameters:

{action: "navigate", payload: "https://example.com"}

CSS and XPath Support

Both selector types supported (XPath auto-detected):

{action: "click", selector: "button.submit"}
{action: "click", selector: "//button[text()='Submit']"}

Auto-Starting Chrome

Browser launches automatically on first use, no manual setup.

Multi-Tab Management

Control multiple tabs with tab_index parameter:

{action: "click", tab_index: 2, selector: "a.email"}

Token Efficiency

• Main skill: 1050 words (target: <500 words for frequent skills)
• Total skill: 6092 words across all files
• Progressive loading ensures only relevant content loaded
• Reference files separated by concern

Comparison with Playwright MCP

Use this skill when:

• Working with existing browser sessions
• Need authenticated workflows
• Managing multiple tabs
• Want minimal overhead

Use Playwright MCP when:

• Need fresh isolated instances
• Generating PDFs/screenshots
• Prefer higher-level abstractions
• Complex automation with built-in retry logic

Credits

Based on superpowers-chrome by obra (Jesse Vincent).

License

MIT