fix(agent): make preview and browser output parsers linear (#7003)

Six regexes over tool output backtracked polynomially on long whitespace
runs or repeated openers: the browser snapshot generic-line filter and
post-interaction split, the task-row and skill-bullet parsers, the email
"no results" check, and the Google /sorry/ interstitial check.

Rewrite each without overlapping quantifiers, keeping the same matches for
the text these tools emit. Move the /sorry/ check into
is_google_sorry_url() so it can be tested.

src/browser_observation.py

@@ -25,7 +25,7 @@ def compact_browser_observation(value, budget=8000):
                 # Snapshot text already contains labels and refs in DOM order.
                 # The refs mapping repeats them and buries menus in raw JSON.
                 lines = [line for line in snapshot.splitlines()
-                         if not re.fullmatch(r'\s*-?\s*generic(?:\s+\[ref=e\d+\])?:?\s*', line)]
+                         if not re.fullmatch(r'\s*(?:-\s*)?generic(?:\s+\[ref=e\d+\])?:?\s*', line)]
                 pages.append(('URL: ' + str(url) + '\n' if url else '') + '\n'.join(lines))
             elif url:
                 notices.append('URL: ' + str(url))
@@ -37,7 +37,8 @@ def compact_browser_observation(value, budget=8000):
                 parsed = json.loads(item)
             except (ValueError, TypeError):
                 # CLI status text precedes the JSON post-interaction state.
-                parts = re.split(r'\n\n\[(?:post-[^\]]+|page state after failed [^\]]+)\]\n', item)
+                # Labels are one line; stopping at \n keeps the split linear.
+                parts = re.split(r'\n\n\[(?:post-[^\]\n]+|page state after failed [^\]\n]+)\]\n', item)
                 if len(parts) > 1:
                     for part in parts:
                         visit(part)

src/clean_agent_preview.py

@@ -1229,7 +1229,7 @@ def tasks_terminal_response(raw, *, user_text='', max_items=20):
     text = str(payload or '').strip()
     rows = []
     for line in text.splitlines():
-        match = re.match(r'^\s*\d+\.\s+(.*?)\s+\(([^)]+)\)\s+—\s+(.+)$', line)
+        match = re.match(r'^\s*\d+\.\s(.*?)\s\(([^()]+)\)\s+—\s+(.+)$', line)
         if match:
             details = match[3].strip()
             rows.append({
@@ -1346,7 +1346,7 @@ def skills_terminal_response(raw, *, user_text='', max_items=20):
             continue
         match = re.match(r'^\s*-\s+\*\*([^*]+)\*\*(?:\s+\(([^)]+)\))?', line)
         if not match:
-            match = re.match(r'^\s*-\s+([^:]+?)(?:\s+\(([^)]+)\))?(?:\s*:|$)', line)
+            match = re.match(r'^\s*-\s+([^\s:][^:]*?)(?:\s\(([^()]+)\)(?:\s*:|$)|:|$)', line)
         if match and status:
             rows.append((status, match.group(1).strip(), (match.group(2) or '').strip()))
     if not rows:
@@ -5065,7 +5065,7 @@ def email_search_result_empty(value):
         try:
             decoded = json.loads(value)
         except (TypeError, ValueError):
-            return bool(re.fullmatch(r'\s*No emails matched [^\n]+\.?\s*', value, re.I))
+            return bool(re.fullmatch(r'\s*No emails matched [^\n]+(?:\n\s*)?', value, re.I))
         if decoded == value:
             return False
         value = decoded
@@ -5339,6 +5339,16 @@ def private_browser_effective_url(result):
     return urls[-1] if urls else ''
 
 
+def is_google_sorry_url(url):
+    """Recognize Google's rate-limit interstitial."""
+    # Matching the host separately keeps this linear; one combined pattern
+    # rescans the host from every dot.
+    return any(
+        re.fullmatch(r'(?:[^/]+\.)?google\.[^/]+', host[1], re.I)
+        for host in re.finditer(r'https?://([^/]+)/sorry/', url, re.I)
+    )
+
+
 def browser_observation_page_missing(raw):
     """Recognize a rendered error page, rather than an article mentioning 404."""
     text = str(raw or '')
@@ -8213,11 +8223,7 @@ async def stream_preview(*, endpoint_url, model, messages, headers, turn_contrac
                         browser_search_url = requested_browser_url or effective_browser_url
                         is_search_engine_navigation = contains_search_engine_navigation(
                             browser_search_url
-                        ) or bool(re.search(
-                            r'https?://(?:[^/]+\.)?google\.[^/]+/sorry/',
-                            effective_browser_url,
-                            re.I,
-                        ))
+                        ) or is_google_sorry_url(effective_browser_url)
                         has_native_search = any(
                             canonical(schema['function']['name']) == 'web_search'
                             for schema in offered