> ## Documentation Index
> Fetch the complete documentation index at: https://docs.webrayn.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Protected Websites

> How webrain handles Cloudflare, Turnstile, CAPTCHA, and authenticated sites — persistent profile + real Chrome + session, native login, and verification rules (no Python sidecar).

# Protected Websites

A protected website is a page state, not a dead end. Webrain's model for gated
or authenticated sites is:

```
persistent profile → real Chrome → persistent session → navigate
  → inspect page state → handle challenge/auth → verify → extract
```

## Rules

1. **Protected navigation starts with real Chrome** + a persistent profile +
   a session. Never the anonymous "blocked → fresh browser → retry" loop.
2. **Read `challenge` on every navigate** (`null` | `cloudflare_challenge` |
   `blocked` | `captcha`). A challenge page is not successful navigation — never
   extract it as target content.
3. **Use the native path first**: `webrain_session(op=login, service, profile, url)`
   (vault + TOTP) or `webrain launch` + `webrain login`. It auto-waits challenges
   and claims Turnstile/reCAPTCHA/hCaptcha widgets; a 2FA/approval gate returns
   `waiting_for_human:true` and the human acts in the headed browser.
4. **Re-attach, don't restart**: an already-authenticated Chrome is reused by
   pointing `CDP_URL` at its port.
5. **No Python sidecar.** Challenge handling is native (vault + TOTP); interactive
   CAPTCHAs the native path can't claim need a human in the headed browser.
6. **Verify before returning**: confirm the target content loaded (not a
   challenge/login/consent page).

## Capability (runtime-dependent)

* obscura / lightpanda cannot pass interactive challenges (the challenge JS crashes).
  Non-interactive Turnstile may pass (verified).
* Real Chrome + profile: native login with vault + TOTP. Challenge handling is an
  internal webrain capability and continues to grow inside the runtime (detect →
  classify → execute → verify); it is not delegated to an external sidecar.

## What NOT to do

* Start protected workflows statelessly.
* Discard a working profile or session after a challenge.
* Switch to a fresh browser after a challenge.
* Claim a bypass succeeded without verifying the target content.
* Treat a challenge/login/consent page as scraped content.

Full anti-pattern list: [Agent Decision Guide](/guides/agent-decision-guide).