Browser Pilot

Universal Browser Pilot Skill

You are equipped with the Universal Browser Pilot Skill. This skill enables you to control a live Chrome browser via the Chrome DevTools MCP server (chrome-devtools-mcp) for any web interaction task—including dynamic web scraping, mass form filling, web app debugging, responsiveness auditing, and persistent web session management.

Capabilities & Toolset

When interacting with web pages via Chrome DevTools MCP, utilize the following core tool patterns:

1. Smart Form Filling & Multi-Input Automation

• Batch Form Filling (fill_form): ALWAYS prefer fill_form over individual fill or click calls when populating multi-field web forms (login screens, registration portals, surveys, enterprise forms). It executes in a single round-trip, preventing state desynchronization.
• Key Navigation & Presses (press_key, type_text): Use press_key for keyboard navigation (e.g. Tab, Enter, Control+Shift+R, Escape) or shortcuts when interacting with custom web controls.
• File Uploads (upload_file): Provide local file paths (e.g., PDFs, invoices, images from /input/ or workspace paths) to target file input elements or file choosers.

2. Dynamic Web Content Extraction & Scraping

• Page Snapshot (take_snapshot): Use take_snapshot to extract structured text representations and unique element uids from single-page applications (React, Vue, Angular, Next.js) where raw HTML scraping fails.
• Structured Export: Extract tables, lists, pricing data, or article text and format them into structured JSON, CSV, or Markdown files saved inside the workspace root /output/ folder (e.g., /output/scraped_data.json).

3. Visual & Runtime Quality Auditing

• Responsive Layout Inspection (take_screenshot, emulate): Test responsive design adherence by emulating mobile viewports vs desktop viewports. Capture visual screenshots and save them directly to /output/screenshot_[device]_[timestamp].png.
• Console Log & Error Capture (list_console_messages, get_console_message): Monitor runtime console logs for unhandled JavaScript exceptions, network failures, or API breaks during automated interaction.
• Performance Tracing (performance_start_trace, performance_stop_trace): Capture execution traces to diagnose rendering bottlenecks or slow network requests.

4. Session Persistence & Authentication Management

• Cookies & Storage Saving: Save session cookies and token states under /output/session_[domain].json after manual or automated login to avoid repetitive login/2FA challenges in subsequent sessions.
• Multi-Tab Orchestration (list_pages, new_page, select_page, close_page): Handle complex workflows requiring multiple tabs or popup windows cleanly.

Universal Operational Workflows

Workflow A: Dynamic Web Scraping & Data Extraction

1. Navigate: Call navigate_page with the target URL.
2. Wait: Use wait_for to ensure dynamic JavaScript content or AJAX requests complete loading.
3. Snapshot: Execute take_snapshot to retrieve content and structural UIDs.
4. Transform: Process extracted data into clean JSON/CSV/Markdown.
5. Persist: Write the resulting deliverable file to output/[filename].[ext].

Workflow B: Mass Form Entry & Document Upload

1. Prepare Data: Read the source data file (JSON, CSV, or parsed document).
2. Inspect Form: Call take_snapshot to identify all form input uids.
3. Fill Mass Form: Call fill_form with an array of mapped values (uid + value).
4. Upload Attachments (If applicable): Call upload_file with the path to any target document.
5. Verify & Execute: Review filled inputs, execute the submit action, and capture a proof screenshot saved to output/submission_proof_[timestamp].png.

Safeguards & Human-in-the-Loop Protocol

1. Dry-Run Mode for High-Risk Actions:
   • Before executing irreversible online actions (e.g., submitting financial payments, deleting data, posting public messages, or final form submissions on live production systems), present a preview summary to the user and request explicit confirmation.
2. Handling CAPTCHA, OTP & 2FA Challenges:
   • If an automated site presents a CAPTCHA or 2FA/OTP screen, pause execution and inform the user. Allow the user to complete the challenge in the visible browser instance, then resume execution automatically upon user signal.
3. Credential Guardrails:
   • Never hardcode login credentials in prompt scripts or skill files. Read credentials securely from local .env files or prompt inputs.