Login Flows

1. Identify Selectors#

Inspect the login page (Headful debugging is great for this). Find:

• Username input (e.g., #username, input[name="email"])
• Password input (e.g., #password)
• Submit button (e.g., button[type="submit"])
• Success indicator (e.g., .dashboard, a[href="/logout"])
• Error message (e.g., .alert-danger)

2. Build the Task#

Create a new Agent task.

Step 1: Navigate#

• Action: Navigate (navigate)
• Value: https://example.com/login

Step 2: Wait for Load#

• Action: Wait for Element (wait_selector)
• Selector: #username (Ensure inputs are ready)

Step 3: Input Credentials#

• Action: Type (type)
  • Selector: #username
  • Value: {$user} (Use a variable!)
• Action: Type (type)
  • Selector: #password
  • Value: {$pass}

Step 4: Submit#

• Action: Click (click)
  • Selector: button[type="submit"]

Step 5: Verification (Crucial!)#

• Action: Wait (wait)
  • Value: 2 (Allow redirect time)
• Action: Condition (if)
  • Condition: exists('.dashboard')
• Inside Condition:
  • Action: Stop (stop)
    • Value: success
• Else:
  • Action: Stop (stop)
    • Value: failure (Login failed)
• End Condition

3. Handling 2FA / CAPTCHA#

If the site uses CAPTCHA or 2FA, fully automated login might be impossible.

• Option A: Use Headful Debugging to log in manually once. Cookies are saved to storage_state.json. Subsequent tasks reuse this session.
• Option B: Use an external CAPTCHA solving service (requires custom JS integration).

4. Reusing the Session#

Once logged in, any subsequent task (even a different one) will use the saved cookies, provided statelessExecution is false.