Summary

Fixes #4631

After a scroll action, the enhanced_dom_tree cache in DOMWatchdog was not being invalidated. This caused subsequent extract() calls to return stale DOM data from before the scroll, leading agents to see duplicate/unchanged content and incorrectly conclude that no new data was available after scrolling.

Problem

Step 1: extract() → builds DOM → caches as enhanced_dom_tree ("18 posts")
Step 2: scroll(down=1) → browser scrolls, new content loads ✅
         → BUT cache NOT cleared! ❌
Step 3: extract() → returns cached OLD data ("18 posts") ← BUG

The agent then sees "duplicate data" → thinks "cannot get more" → may give up on pagination.

Root Cause

The scroll action in tools/service.py never called dom_watchdog.clear_cache().

Other operations that correctly clear the cache:

• read_file (source=page) — service.py:1736
• Tab switch (AgentFocusChangedEvent) — session.py:1141

The cache read path at markdown_extractor.py:119-125 returns cached value without any staleness check.

Fix

Add dom_watchdog.clear_cache() call after successful scroll completion:

# After successful scroll, clear DOM cache to ensure next extract gets fresh content
if browser_session._dom_watchdog:
    browser_session._dom_watchdog.clear_cache()

Changes

• browser_use/tools/service.py: Added 3 lines — clear_cache() call after successful scroll return

Verification

• ✅ ruff check passes
• ✅ ruff format passes
• ✅ pyright passes (0 errors, 0 warnings)

Summary by cubic

Clear the DOMWatchdog enhanced_dom_tree cache after a successful scroll so the next extract() reads fresh DOM. This prevents stale or duplicate content after pagination.

• Bug Fixes
  • In browser_use/tools/service.py, call browser_session._dom_watchdog.clear_cache() on successful scroll.
  • Ensures extract() doesn’t return pre-scroll cached DOM.

^{Written for commit 4acd164. Summary will update on new commits.}