fix(captcha): detect Cloudflare challenge pages

[alectimison-maker]

Summary

This is the third and final planned follow-up from #505, after #2720 and #2721. It adds a browser-level detector for full-page Cloudflare managed Challenge Pages and routes them into the existing manual CAPTCHA gate.

The branch now includes current main through #2721. The integration preserves the canonical CAPTCHA matcher and token normalization from main while keeping the dedicated Cloudflare response signal authoritative during restored-state hydration.

Cloudflare documents cf-mitigated: challenge as the stable response signal for Challenge Pages. The implementation therefore arms only from that header on a top-level main_frame response. Requests under /cdn-cgi/challenge-platform/ are deliberately insufficient on their own because Cloudflare also uses that endpoint for general bot-management JavaScript and embedded widgets.

What changed

• Observe top-level response headers in both Chrome and Firefox and arm a manual gate only for cf-mitigated: challenge.
• Treat challenge-platform requests as supporting diagnostics only after a response-backed gate is active.
• Keep the signal independent of English dialog text, so localized or script-rendered interstitials remain blocked.
• Persist a sanitized origin + pathname document identity in storage.session, without query strings, fragments, response headers, or page content, so a service-worker restart cannot fail open.
• Block mutation and automatic solve_captcha dispatch while allowing explicit navigation away from the challenged document.
• Clear the gate on a subsequent normal top-level response, navigation to a different document, or tab cleanup. Clearing the chat alone intentionally preserves the active tab-level guard.
• Keep the state-transition module and behavior tests symmetric across the Chrome and Firefox builds.

References:

• https://developers.cloudflare.com/cloudflare-challenges/challenge-types/challenge-pages/detect-response/
• https://developers.cloudflare.com/fundamentals/reference/cdn-cgi-endpoint/
• https://developers.cloudflare.com/cloudflare-challenges/challenge-types/challenge-pages/

Tests

• node test/run.js — 1586/1586 passed.
• npm run test:toolbar-guard — 33/33 passed.
• npm run test:security — 60/60 passed.
• Chrome and Firefox Cloudflare state modules are byte-identical.
• GitHub Actions smoke — passed on a6bd6b82.

The regression coverage includes header casing, XHR exclusion, Turnstile/challenge-platform false-positive prevention, query-string redaction, localized-page retention, mutation/solver blocking, navigation abandonment, worker restart hydration, restart-first response/navigation clearing, concurrent hydration blocking, serialized storage updates, authoritative stale-snapshot rejection, chat clearing, tab cleanup, and Chrome/Firefox listener parity.

[vercel]

@alectimison-maker is attempting to deploy a commit to the esokullu's projects Team on Vercel.

A member of the Team first needs to authorize it.

[Copilot]

Pull request overview

Adds browser-level Cloudflare Challenge Page detection and routes detected interstitials through the manual CAPTCHA gate.

Changes:

• Detects cf-mitigated: challenge on top-level responses.
• Persists sanitized challenge state across worker restarts.
• Adds symmetric Chrome/Firefox integration and regression tests.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
test/run.js	Tests detection, persistence, gating, cleanup, and parity.
src/chrome/src/background.js	Wires Chrome network and navigation listeners.
src/firefox/src/background.js	Wires Firefox network and navigation listeners.
src/chrome/src/agent/cloudflare-managed-challenge.js	Implements Chrome challenge state transitions.
src/firefox/src/agent/cloudflare-managed-challenge.js	Implements Firefox challenge state transitions.
src/chrome/src/agent/agent.js	Integrates persistence and manual gating in Chrome.
src/firefox/src/agent/agent.js	Integrates persistence and manual gating in Firefox.

Suppressed comments (2)

src/chrome/src/agent/agent.js:7967

• The Cloudflare gate is also persisted inside the conversation snapshot, but clearing the dedicated signal does not rewrite that snapshot. After a normal response clears the live gate and a later worker restart finds no dedicated signal, this line restores the stale conversation copy as an active Cloudflare gate; future normal responses cannot clear it because _cloudflareManagedChallenges is empty. Treat the dedicated Cloudflare storage key as authoritative and never restore its gate from captchaGateState.
  src/firefox/src/agent/agent.js:984
• The Cloudflare gate is also persisted inside the conversation snapshot, but clearing the dedicated signal does not rewrite that snapshot. After a normal response clears the live gate and a later background restart finds no dedicated signal, this line restores the stale conversation copy as an active Cloudflare gate; future normal responses cannot clear it because _cloudflareManagedChallenges is empty. Treat the dedicated Cloudflare storage key as authoritative and never restore its gate from captchaGateState.

          if (!cloudflareSignal) this._captchaGateStates.set(tabId, captchaGateState);

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[alectimison-maker]

Addressed the Copilot review in 50106d10:

• hydrate and serialize response/navigation/platform transitions per tab, including restart-first clear events;
• share one in-flight hydration promise so concurrent callers remain fail-closed;
• make the dedicated Cloudflare storage key authoritative, omit its gate from conversation snapshots, and reject legacy stale copies (including the suppressed review concern);
• add Chrome/Firefox regressions for restart ordering, concurrent hydration, serialized storage writes, and stale snapshot rejection.

Validation: 1571 unit tests passed (one unrelated upstream changelog failure remains), security 60/60, toolbar guard 33/33, and the new-head WebMCP E2E check passed.