Fix frame-filter parsing for host-port urls

Co-authored-by: Shri Sukhani <shrisukhani@users.noreply.github.com>

Author: cursoragent

currentState.md

@@ -142,6 +142,7 @@ HyperAgent exposes a TypeScript SDK for browser automation with three primary pa
 - Added an additional open-tab fallback path: when the tab array becomes unreadable (e.g. trapped `length`), prompt assembly now still emits the current tab line instead of a blank/no-tabs summary.
 - Added constructor regression coverage for trap-prone `llm` config getters, ensuring fallback failure paths stay deterministic and readable.
 - Hardened CDP frame-filter URL normalization to support protocol-relative and scheme-less frame URLs while avoiding path-only false positives in host-based ad-domain detection.
+- Refined CDP frame-filter URL normalization to correctly handle scheme-less `host:port` URLs (without misclassifying them as custom schemes), preserving ad-domain detection coverage in those cases.
 - Hardened prompt base-message materialization with trap-safe array reads so malformed/trap-prone seed message arrays no longer crash message assembly and readable entries are preserved.
 - Hardened constructor custom-action ingestion with trap-safe array reads so unreadable custom-action entries are skipped while valid entries continue to register.
 - Expanded top-level package exports for key workflow/config types at `@hyperbrowser/agent`.

src/cdp/frame-filters.test.ts

@@ -42,6 +42,14 @@ describe("isAdOrTrackingFrame", () => {
     ).toBe(true);
   });
 
+  it("filters known ad domains with host:port urls missing explicit protocol", () => {
+    expect(
+      isAdOrTrackingFrame({
+        url: "securepubads.g.doubleclick.net:443/pagead/ads",
+      })
+    ).toBe(true);
+  });
+
   it("filters obvious pixel-style tracking frames", () => {
     expect(
       isAdOrTrackingFrame({

src/cdp/frame-filters.ts

@@ -101,7 +101,17 @@ const TRACKING_PARAMS = [
 
 const MIN_FILTER_SCORE = 2;
 
-const HAS_PROTOCOL_PATTERN = /^[a-z][a-z0-9+\-.]*:/i;
+const HAS_SCHEME_WITH_SLASHES_PATTERN = /^[a-z][a-z0-9+\-.]*:\/\//i;
+const SPECIAL_SCHEME_PREFIXES = [
+  "about:",
+  "data:",
+  "blob:",
+  "file:",
+  "javascript:",
+  "mailto:",
+  "chrome:",
+  "devtools:",
+];
 
 function normalizeUrlForParsing(value: string): string {
   const trimmed = value.trim();
@@ -111,7 +121,12 @@ function normalizeUrlForParsing(value: string): string {
   if (trimmed.startsWith("//")) {
     return `https:${trimmed}`;
   }
-  if (HAS_PROTOCOL_PATTERN.test(trimmed)) {
+  if (
+    HAS_SCHEME_WITH_SLASHES_PATTERN.test(trimmed) ||
+    SPECIAL_SCHEME_PREFIXES.some((prefix) =>
+      trimmed.toLowerCase().startsWith(prefix)
+    )
+  ) {
     return trimmed;
   }
   if (trimmed.startsWith("/")) {