Singapore Financial Balance Checker

Extract balances and transaction data from Singapore financial websites via browser automation.

Workflow

Prerequisites

Ensure you have the following installed:

• Bun runtime (install via curl -fsSL https://bun.sh/install | bash)
• Agent-browser (install via npm install -g agent-browser)
• Chromium (install via agent-browser install)
• Read history folder for latest information and trouble shooting tips

Steps

# 1. Open financial site (see references/supported-websites.md for URLs)
npx agent-browser open https://app.sg.endowus.com/ --headed

# 2. USER LOGS IN MANUALLY (Singpass, 2FA, etc.)
# Wait for user confirmation before proceeding

# 3. Navigate to portfolio/accounts
npx agent-browser snapshot -i
npx agent-browser find text "Portfolio" click
npx agent-browser wait 2000

# 4. Extract data
npx agent-browser snapshot -c
npx agent-browser eval "document.body.innerText" --json

# 5. Close when done
npx agent-browser close

Key Commands

Authentication

CRITICAL Security Rules:

• Always use --headed flag so user can see and interact
• NEVER ask for or handle credentials programmatically
• NEVER store or log passwords, OTPs, or 2FA codes
• User must log in manually in the browser window
• Many SG sites require Singpass/2FA - user must complete this themselves

Handling Singpass Login

1. Wait for Singpass QR or redirect
2. User scans with Singpass app or enters credentials
3. User completes 2FA on their phone
4. Wait for redirect: npx agent-browser wait --url "**/dashboard**"

Handling OTP / TOTP

1. Wait for the user to login
2. Once on the OTP page, ask the user to enter the OTP into the browser

Data Extraction

Common Data Points

Transaction/Activity History

Extract Transaction Table

npx agent-browser eval "Array.from(document.querySelectorAll('table')[1].querySelectorAll('tbody tr')).map(row => { const cells = row.querySelectorAll('td'); return cells.length >= 4 ? [cells[0].innerText.trim(), cells[1].innerText.replace(/\\n/g, ' ').trim(), cells[2].innerText.trim(), cells[3].innerText.trim()].join(' | ') : null; }).filter(x => x && !x.includes('Load more')).join('\\n')"

Pagination Handling

# Load all transactions
for i in {1..15}; do
  npx agent-browser find text "Load more" click 2>/dev/null || break
  npx agent-browser wait 800
done

# Verify all loaded (should show "1-194 of 194")
npx agent-browser eval "document.body.innerText.match(/\\d+-\\d+ of \\d+/)?.[0]"

Security Best Practices

DO:

• Always use --headed mode for financial sites
• Let users authenticate manually (Singpass, OTP, 2FA)
• Close browser sessions after use
• Verify you're on the correct domain before proceeding

DON'T:

• Never capture, store, or transmit passwords
• Never ask the user to tell you their password
• Never screenshot login pages or OTP screens
• Never automate credential entry
• Never store session cookies for financial sites

Error Handling

Verification Checklist

Before completing extraction:

• Browser opened with --headed flag
• User completed authentication manually
• Navigated to correct account/portfolio page
• All paginated data loaded (if applicable)
• Data extracted and validated
• Browser closed with npx agent-browser close

Scripts

Reusable shell scripts in scripts/ for common automation tasks.

Dealing with Dynamic Sites and Documenting new lines

• It's common for financial websites to always update their page design, layout, security system and more.
• What works today may not work tomorrow.
• Everytime you discover a way to fetch information that the user is looking for, update it in history folder within the skills directory.
• Create a new file for each institution, documenting the changes and the new approach. (e.g dbs.md)
• Be very specific about the new approach that you discovered

Additional Resources

• references/supported-websites.md - Full list of institutions and login URLs
• references/when-snapshot-returns-empty.md - For DBS/OCBC legacy iframe sites