Kuang Grade Mark Eleven: CAPTCHA, Anti-Bot & HIL Strategies
1. Challenge Triage Matrix
When an agent encounters a block or challenge page in Kuang Browser, it MUST classify the challenge before acting:
| Challenge Type | Appearance / Indicators | Recommended Autonomous Action | Escalation (HIL) |
|---|---|---|---|
| Simple Text CAPTCHA | Distorted letters/numbers image + text input box (name="captcha", id="captcha_input") |
Read text from image via take_screenshot(marks:true) or observe(as="markdown"), type text into box with act(mark_id, action="type", text="…"), click submit. |
If 2 failed attempts, trigger HIL. |
| Math / Logic Challenge | Simple text e.g. "What is 4 + 7?" or "Type the 3rd word" | Evaluate the math expression in prompt, type answer into input. | None needed. |
| Cloudflare Turnstile Checkbox | "Verify you are human" iframe box | Try clicking the Turnstile checkbox via act(mark_id, action="click"). |
If Turnstile enters infinite loop or shows interactive puzzle, trigger HIL. |
| Complex Puzzle / Geetest / reCAPTCHA v2 Image Grid | "Select all traffic lights" / slider puzzle | DO NOT GUESS. These burn time and flag the IP. | Trigger request_human_intervention or send_notification IMMEDIATELY. |
| 2FA / SMS OTP / Login Prompt | "Enter code sent to +1-XXX..." or password prompt | STOP. The agent does not have access to the user's phone or 2FA authenticator app. | Trigger request_human_intervention IMMEDIATELY. |
2. Triggering Human-In-The-Loop (HIL)
When human intervention is required:
Option A: Blocking Intervention (request_human_intervention)
Halts the agent loop, plays an audible alert, brings the Kuang browser window to the front, and presents a HUD banner on screen until the human solves the challenge and clicks "Resolved":
{
"reason": "Cloudflare Turnstile verification required to access target page",
"timeout_seconds": 180,
"bring_to_front": true
}
Option B: Non-Blocking Desktop Alert (send_notification)
If you want to notify the user via a native Windows toast notification without completely freezing execution:
{
"title": "Kuang Browser: Human Intervention Required",
"message": "Please complete SMS 2FA verification in the active browser window.",
"urgency": "critical"
}