From 1e68aa5d7bb9395b13a3cd9ceac4b358bdb5f303 Mon Sep 17 00:00:00 2001
From: ciregenz <ciregenz@berkeley.edu>
Date: Wed, 15 Jul 2026 11:54:49 -0700
Subject: [PATCH] [eric] browser: send-script completes on X/IG/FB
 (submit-button vocab beyond LinkedIn Send)

---
 backend/apps/agents/browser/browser_agent.py | 17 ++++++++++++++---
 backend/tests/test_browser_agent_loop.py     | 14 ++++++++++++++
 backend/tests/test_browser_send_script.py    | 10 ++++++++--
 3 files changed, 36 insertions(+), 5 deletions(-)

diff --git a/backend/apps/agents/browser/browser_agent.py b/backend/apps/agents/browser/browser_agent.py
index a51bc388..a0294799 100644
--- a/backend/apps/agents/browser/browser_agent.py
+++ b/backend/apps/agents/browser/browser_agent.py
@@ -473,13 +473,24 @@ def delta_state(text: str, seen_lines: set[str]) -> str:
 # A button row whose name is exactly a Send control (not "Send InMail credit" or "Send a message to X"); used to hand the model the Send button after it types, so it never burns turns hunting a button that's right there.
 P_SEND_ROW_RE = re.compile(r'\[(\d+)\]\*?<\s*button\s+"([^"]*)"', re.I)
 
+# The submit-button labels across the popular composers, so the fast send-path COMPLETES on
+# X/IG/FB/Threads/YouTube, not just LinkedIn. EXACT match + button-only (P_SEND_ROW_RE) is the
+# safety: "Post" never matches "Post a job", a "Share" menuitem isn't a <button so it's excluded,
+# and every caller is post-fill (a hint, or the send-script's receipt-gated click), so a rare
+# opener-vs-submit mismatch fails safe (the composer just doesn't clear), never a false send.
+P_SEND_LABELS = frozenset({
+    "send", "send now", "send message",              # LinkedIn / Gmail / DMs
+    "post", "post all", "tweet", "reply",            # X / Threads compose + reply
+    "publish", "comment", "share",                   # articles / YouTube+FB comments / shares
+})
+
 
 def send_index_in_state(state_text: str):
-    """(index, name) of a real Send button in an interactives list, or None.
-    Strict exact match so it never grabs an upsell or a profile 'Send a message' link."""
+    """(index, name) of a real submit button in an interactives list, or None. Strict exact
+    match against P_SEND_LABELS so it never grabs an upsell or a profile 'Send a message' link."""
     for line in (state_text or "").splitlines():
         m = P_SEND_ROW_RE.search(line)
-        if m and m.group(2).strip().lower() in ("send", "send now", "send message"):
+        if m and m.group(2).strip().lower() in P_SEND_LABELS:
             return int(m.group(1)), m.group(2)
     return None
 
diff --git a/backend/tests/test_browser_agent_loop.py b/backend/tests/test_browser_agent_loop.py
index 778cf806..2cc9eaf6 100644
--- a/backend/tests/test_browser_agent_loop.py
+++ b/backend/tests/test_browser_agent_loop.py
@@ -1524,6 +1524,20 @@ def test_send_index_handoff_points_only_at_a_real_send_button():
     assert send_index_in_state("") is None
 
 
+def test_send_index_finds_the_popular_composers_submit_buttons():
+    # the generalization: the fast send-path must COMPLETE on X/IG/FB/YouTube, so the submit
+    # finder knows Post/Reply/Tweet/etc, not just LinkedIn's "Send".
+    from backend.apps.agents.browser.browser_agent import send_index_in_state
+    assert send_index_in_state('[3]<textbox "Post your reply">\n[8]<button "Reply">') == (8, "Reply")
+    assert send_index_in_state('[2]<textbox "What is happening?">\n[9]<button "Post">') == (9, "Post")
+    assert send_index_in_state('[4]<button "Tweet">') == (4, "Tweet")
+    assert send_index_in_state('[7]<button "Comment">') == (7, "Comment")
+    # exact + button-only keeps the safety: not a look-alike, not a menuitem, not an opener phrase
+    assert send_index_in_state('[5]<button "Post a job">') is None
+    assert send_index_in_state('[6]<menuitem "Share">') is None          # button-only excludes the Drive-Share false positive
+    assert send_index_in_state('[8]<button "Reply to Maya">') is None
+
+
 def test_strip_lone_surrogates():
     from backend.apps.agents.browser.browser_agent import strip_lone_surrogates, format_tool_result
     # an orphan UTF-16 surrogate (half an emoji from the webview) is what crashes the turn at .encode('utf-8'); it must be swapped, not carried through
diff --git a/backend/tests/test_browser_send_script.py b/backend/tests/test_browser_send_script.py
index bd19ec9f..670d564e 100644
--- a/backend/tests/test_browser_send_script.py
+++ b/backend/tests/test_browser_send_script.py
@@ -69,8 +69,11 @@ async def test_surface_gate_declines_when_no_composer_in_perception():
 
 @pytest.mark.asyncio
 async def test_surface_gate_fires_on_a_NON_linkedin_composer():
-    """The whole generalization: the same send-script fires on ANY site whose
-    perception carries a composer (here X's 'Post your reply'), no per-site URL gate."""
+    """The whole generalization: the same send-script fires on ANY site whose perception
+    carries a composer (here X's 'Post your reply'), no per-site URL gate, AND completes
+    by clicking X's real submit button ('Reply') BY INDEX. Before the submit-vocabulary
+    was generalized this only 'passed' because the mock succeeds on the by-name 'Send'
+    fallback that doesn't exist on real X, so assert the real index path is taken."""
     X_FILLED = '[3]<textbox "Post your reply" value="[test] hello world r9-os">\n[8]<button "Reply">'
     X_SENT = '[3]<textbox "Post your reply">\n[1]<link "Home">'
     ex, calls = make_exec([X_FILLED, X_FILLED, X_SENT])
@@ -78,6 +81,9 @@ async def test_surface_gate_fires_on_a_NON_linkedin_composer():
                                  payload_in_textbox, payload_source=TASK,
                                  current_url="https://x.com/messages/123")
     assert r is not None and r["sent"] is True
+    # the send went via BrowserClickIndex on X's real Reply submit (8), not the by-name 'Send' crutch
+    assert ("BrowserClickIndex", {"index": 8}) in calls["clicks"]
+    assert not any(t == "BrowserClickByName" for t, _ in calls["clicks"])
 
 
 @pytest.mark.asyncio