LM Arena Bridge

When using lmarenabridge, you may need to implement a Cloudflare/Turnstile pass-through logic within the get_recaptcha_v3_token function in main.py to bypass security challenges on https://arena.ai/.

To implement this:

1. Ensure AsyncCamoufox is initialized with headless=False and main_world_eval=True.
2. Navigate to https://arena.ai/ using domcontentloaded wait state.
3. Implement a loop to check for the "Just a moment" page title or the presence of the Turnstile widget.
4. Use the click_turnstile(page) function to interact with the challenge if detected.
5. Remove any existing logic that specifically waits for the arena-auth-prod-v1 cookie, as the priority is passing the challenge to allow the grecaptcha library to load.

# Step 1: Ensure browser initialization is set to non-headless
async with AsyncCamoufox(headless=False, main_world_eval=True) as browser:

# Step 2: Implement Challenge Logic inside get_recaptcha_v3_token
debug_print("  🌐 Navigating to arena.ai...")
await page.goto("https://arena.ai/", wait_until="domcontentloaded")

# --- NEW: Cloudflare/Turnstile Pass-Through ---
debug_print("  🛡️  Checking for Cloudflare Turnstile...")

try:
    for _ in range(5):
        title = await page.title()
        if "Just a moment" in title:
            debug_print("  🔒 Cloudflare challenge active. Attempting to click...")
            clicked = await click_turnstile(page)
            if clicked:
                debug_print("  ✅ Clicked Turnstile.")
                await asyncio.sleep(3)
        else:
            await click_turnstile(page)
            break
        await asyncio.sleep(1)
    
    await page.wait_for_load_state("domcontentloaded")
except Exception as e:
    debug_print(f"  ⚠️ Error handling Turnstile: {e}")
# ----------------------------------------------

The bridge can optionally use an AI-powered solver to handle reCAPTCHA v2 image challenges using YOLO models. This is useful when standard reCAPTCHA v3 scoring fails.

Installation Install the bridge with the [vision] extra or install the solver directly:

pip install -e ".[vision]"
# or
pip install vision-ai-recaptcha-solver

Configuration To enable the solver, add the following to your config.json:

{
  "vision_recaptcha_solver": {
    "enabled": true,
    "headless": true,
    "timeout": 180,
    "max_attempts": 3,
    "browser_path": "C:\\path\\to\\chrome.exe",
    "proxy": "http://user:pass@ip:port"
  }
}

Requirements

• Python 3.10+
• A Chromium-based browser (Chrome for Testing, Edge)
• The vision-ai-recaptcha-solver package (installs heavy dependencies like torch, ultralytics, and onnxruntime)

Note: If the dependency is not installed, the bridge will continue to work normally but will silently skip the vision fallback.

The Userscript Proxy operates by injecting a fetch_script into the browser page. This script intercepts requests and manages complex authentication flows, specifically reCAPTCHA, to ensure requests are valid.

Key behaviors include:

• Automatic reCAPTCHA Minting: If a request payload contains a recaptchaV3Token that is missing or invalid, the script automatically mints a new v3 token using grecaptcha.execute.
• Retry Logic: If an upstream fetch returns a 403 status code containing 'recaptcha', the script attempts to retry once with a fresh v3 token. If that fails, it falls back to minting a reCAPTCHA v2 token.
• Streaming Support: The script uses a ReadableStream reader to process responses. It decodes chunks using TextDecoder and emits newline-delimited lines back to the host via console.log using the prefix LM_BRIDGE_PROXY|{jid}|.
• Debugging: When LM_BRIDGE_PROXY_DEBUG is set, the script emits detailed debug events (e.g., before_fetch, after_fetch, v3_minted) through the same console logging mechanism.

/* The proxy communicates via console.log with a specific format: */
console.log('LM_BRIDGE_PROXY|' + jid + '|' + JSON.stringify(obj));

The bridge manages 403 Forbidden errors differently depending on the transport used:

Userscript Proxy

If the transport is userscript, a 403 is often treated as a transient reCAPTCHA validation issue. The bridge waits for a grace period (configured via userscript_proxy_recaptcha_grace_seconds) to allow the in-page script to retry. If recaptcha validation failed is detected in the response body, it tracks consecutive failures. After 5 failures, it fails with a recaptcha_error.

Browser Transports (Chrome/Camoufox)

If using chrome or camoufox, a 403 triggers a side-channel reCAPTCHA token refresh via refresh_recaptcha_token(force_new=True).

Transport Switching

If consecutive reCAPTCHA failures occur (threshold of 2), the bridge automatically switches transport preference:

• From chrome $\rightarrow$ camoufox (sets use_browser_transports = True and prefer_chrome_transport = False).
• From other browser transports $\rightarrow$ chrome (sets use_browser_transports = True and prefer_chrome_transport = True).