From 99bd15fdb14062fca9ed338e1e5c90fab4d8ad93 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Fri, 5 Jun 2026 11:14:13 -0700
Subject: [PATCH] [eric] browser: attached page state shrinks to changed rows,
 unchanged indices stay valid

---
 backend/apps/agents/browser/browser_agent.py | 36 ++++++++++++++++++-
 backend/tests/test_browser_agent_loop.py     | 38 ++++++++++++++++++++
 2 files changed, 73 insertions(+), 1 deletion(-)

diff --git a/backend/apps/agents/browser/browser_agent.py b/backend/apps/agents/browser/browser_agent.py
index 3eefabfd..e10d5a17 100644
--- a/backend/apps/agents/browser/browser_agent.py
+++ b/backend/apps/agents/browser/browser_agent.py
@@ -148,9 +148,33 @@ def _truncate_state(text: str, max_lines: int = _AUTO_STATE_MAX_LINES) -> str:
     )
 
 
+def _delta_state(text: str, seen_lines: set[str]) -> str:
+    """Shrink an attached element list to the rows that changed since the last
+    attach; stable indices make a line's identity meaningful, so re-sending 30
+    unchanged rows every action is pure token burn. Mutates `seen_lines` to the
+    new baseline. Small overlaps just resend the full list (a reshuffle)."""
+    rows = [l for l in str(text).splitlines() if l.startswith("[")]
+    cur = set(rows)
+    prev = set(seen_lines)
+    seen_lines.clear()
+    seen_lines.update(cur)
+    if not prev or not rows:
+        return text
+    fresh = [l for l in rows if l not in prev]
+    unchanged = len(rows) - len(fresh)
+    if unchanged < 6:
+        return text
+    if not fresh:
+        return f"(all {unchanged} element rows unchanged since your last look; same numbers still valid)"
+    return "\n".join(fresh) + (
+        f"\n(+{unchanged} rows unchanged since your last look; their numbers are still valid)"
+    )
+
+
 async def _post_action_state(
     tool_name: str, tool_input: dict, result: dict,
     browser_id: str, tab_id: str, wait_exec, goal: str,
+    seen_lines: set[str] | None = None,
 ) -> str:
     """Settle the page after a mutating action, then return a compact fresh
     interactives list to append to its result. Empty string = attach nothing."""
@@ -174,7 +198,8 @@ async def _post_action_state(
         return ""
     if not isinstance(lst, dict) or "error" in lst or not lst.get("text"):
         return ""
-    return f"\n\n{PAGE_STATE_MARKER}\n{_truncate_state(lst['text'])}"
+    state = lst["text"] if seen_lines is None else _delta_state(lst["text"], seen_lines)
+    return f"\n\n{PAGE_STATE_MARKER}\n{_truncate_state(state)}"
 
 
 async def _request_browser_approval(
@@ -651,6 +676,8 @@ async def run_browser_agent(
     # MAX_TURNS doing the same broken thing.
     consecutive_violations = 0
     MAX_CONSECUTIVE_VIOLATIONS = 3
+    # rows already shown to the model; attached state shrinks to the delta
+    attached_state_seen: set[str] = set()
     try:
         for turn in range(MAX_TURNS):
             if cancel_event.is_set():
@@ -1154,8 +1181,15 @@ async def run_browser_agent(
                     last_seen_url = result["url"]
                 card_gone_streak = card_gone_streak + 1 if card_is_unavailable(result) else 0
 
+                # a direct full list resets the delta baseline to what the model just saw
+                if tu.name == "BrowserListInteractives" and "error" not in result:
+                    attached_state_seen.clear()
+                    attached_state_seen.update(
+                        l for l in str(result.get("text") or "").splitlines() if l.startswith("[")
+                    )
                 _auto_state = await _post_action_state(
                     tu.name, tu.input, result, browser_id, tab_id, _wait_exec, current_next_goal,
+                    seen_lines=attached_state_seen,
                 )
                 if _auto_state:
                     result["text"] = f"{result.get('text') or ''}{_auto_state}"
diff --git a/backend/tests/test_browser_agent_loop.py b/backend/tests/test_browser_agent_loop.py
index 8cfe6ee2..d8a72d72 100644
--- a/backend/tests/test_browser_agent_loop.py
+++ b/backend/tests/test_browser_agent_loop.py
@@ -1240,3 +1240,41 @@ def test_post_action_state_hung_settle_attaches_nothing(monkeypatch):
         "b1", "", _fake_exec(calls), "",
     ))
     assert out == "" and calls == []
+
+
+def test_delta_state_first_attach_sends_full_list():
+    from backend.apps.agents.browser.browser_agent import _delta_state
+    seen = set()
+    full = "8 interactive elements:\n" + "\n".join(f'[{i}]<button "b{i}">' for i in range(1, 9))
+    assert _delta_state(full, seen) == full
+    assert len(seen) == 8
+
+
+def test_delta_state_shrinks_to_changed_rows():
+    from backend.apps.agents.browser.browser_agent import _delta_state
+    rows = [f'[{i}]<button "b{i}">' for i in range(1, 11)]
+    seen = set()
+    _delta_state("\n".join(rows), seen)
+    nxt = rows[:9] + ['[10]<button "b10" value="typed">', '[11]*<button "new">']
+    out = _delta_state("\n".join(nxt), seen)
+    assert '[11]*<button "new">' in out and 'value="typed"' in out
+    assert '[3]<button "b3">' not in out
+    assert "+9 rows unchanged" in out
+    assert seen == set(nxt)
+
+
+def test_delta_state_no_changes_collapses_to_one_line():
+    from backend.apps.agents.browser.browser_agent import _delta_state
+    rows = "\n".join(f'[{i}]<link "l{i}">' for i in range(1, 13))
+    seen = set()
+    _delta_state(rows, seen)
+    out = _delta_state(rows, seen)
+    assert out.startswith("(all 12 element rows unchanged")
+
+
+def test_delta_state_reshuffle_resends_full():
+    from backend.apps.agents.browser.browser_agent import _delta_state
+    seen = set()
+    _delta_state("\n".join(f'[{i}]<button "a{i}">' for i in range(1, 11)), seen)
+    new_page = "10 interactive elements:\n" + "\n".join(f'[{i}]<button "z{i}">' for i in range(1, 11))
+    assert _delta_state(new_page, seen) == new_page