Stop hiding response which contains only partial verification keywors.

Author: beacoder

python_agent_harness/tui.py

@@ -147,15 +147,20 @@ def _strip_final_check(text: str) -> str:
 
     - an explicit "[FINAL CHECK]" header: everything from it is dropped;
     - the bare checklist: a trailing block of Goal:/Status:/Evidence:
-      bullets with at least 2 distinct labels (fallback for replies
-      that answer the completion rules without the header).
-
-    The agent loop still produces and stores the message unchanged;
-    this only trims it from the TUI display.
+      bullets with ALL THREE labels (fallback for replies that answer
+      the completion rules without the header; requiring all three
+      avoids eating genuine summary bullets like "- Status: ...").
+
+    The strip NEVER empties a reply: if the check block is the whole
+    message (no real content before it), the message is kept intact —
+    hiding it would hide the model's only visible response.  The agent
+    loop still produces and stores the message unchanged; this only
+    trims it from the TUI display.
     """
     idx = text.find("[FINAL CHECK]")
     if idx != -1:
-        return text[:idx].rstrip()
+        prefix = text[:idx].rstrip()
+        return prefix if prefix else text
     lines = text.splitlines()
     labels: set[str] = set()
     first: int | None = None
@@ -166,13 +171,14 @@ def _strip_final_check(text: str) -> str:
         labels.add(label)
         if first is None:
             first = i
-    if len(labels) < 2 or first is None:
+    if len(labels) < len(_FINAL_CHECK_LABELS) or first is None:
         return text
     # only a TRAILING bullet block is the completion check — bullets in
     # the middle of a real reply (followed by more content) stay visible
     if not all(not line.strip() or _is_bullet_line(line) for line in lines[first:]):
         return text
-    return "\n".join(lines[:first]).rstrip()
+    prefix = "\n".join(lines[:first]).rstrip()
+    return prefix if prefix else text
 
 
 def _strip_reasoning(text: str, reasoning: str) -> str:

tests/test_tui.py

@@ -168,8 +168,11 @@ def test_nudge_and_final_check_hidden(self):
         out = buf.getvalue()
         self.assertIn("build the thing", out)
         self.assertIn("Done. All tests pass.", out)
-        self.assertNotIn("FINAL CHECK", out)
-        self.assertNotIn("Status: SUCCESS", out)
+        # the [FINAL CHECK] block on a content-carrying reply is hidden,
+        # but a reply that IS only the check block stays visible — the
+        # model's only visible response must not vanish
+        self.assertNotIn("FINAL CHECK\n- Goal: build the thing", out)
+        self.assertIn("Status: SUCCESS", out)
         self.assertNotIn(config.NUDGE_MESSAGE, out)
         # the agent loop's history is untouched
         self.assertEqual(
@@ -185,7 +188,7 @@ def test_nudge_and_final_check_hidden(self):
     def test_final_check_without_header_stripped(self):
         """A reply that answers the completion rules WITHOUT the
         [FINAL CHECK] header (just the checklist bullets) is hidden too:
-        at least 2 of Goal:/Status:/Evidence: in the trailing block."""
+        all three of Goal:/Status:/Evidence: in the trailing block."""
         tui, buf = make_tui()
         tui.session.last_messages = [
             Message(role="user", content="build the thing"),
@@ -202,9 +205,9 @@ def test_final_check_without_header_stripped(self):
         self.assertNotIn("Status: SUCCESS", out)
         self.assertNotIn("Evidence:", out)
 
-    def test_final_check_pure_bullets_stripped(self):
-        """A pure-checklist reply (no header, no content, "•" bullets)
-        vanishes entirely from the panel."""
+    def test_final_check_pure_bullets_kept(self):
+        """A pure-checklist reply (no header, no other content) stays
+        visible — stripping it would hide the model's only response."""
         tui, buf = make_tui()
         tui.session.last_messages = [
             Message(role="user", content="build the thing"),
@@ -215,23 +218,28 @@ def test_final_check_pure_bullets_stripped(self):
         ]
         tui.console.print(tui._render_conversation())
         out = buf.getvalue()
-        self.assertNotIn("Status", out)
-        self.assertNotIn("Evidence", out)
-
-    def test_regular_reply_mentioning_one_label_kept(self):
-        """A genuine reply mentioning a single checklist label is NOT
-        stripped — at least 2 distinct labels are required."""
+        self.assertIn("Status: SUCCESS", out)
+        self.assertIn("Evidence: done", out)
+        self.assertIn("Goal: build it", out)
+
+    def test_partial_checklist_kept(self):
+        """A genuine reply ending in only two checklist-looking bullets
+        is NOT stripped — all three labels are required for the
+        no-header fallback, so real summary bullets survive (also
+        covers the single-label case, which takes the same path)."""
         tui, buf = make_tui()
         tui.session.last_messages = [
             Message(role="user", content="build the thing"),
             Message(
                 role="assistant",
-                content="Goal: build the thing. Now verifying the build...",
+                content="Deployed.\n- Status: SUCCESS\n- Goal: deploy the service",
             ),
         ]
         tui.console.print(tui._render_conversation())
         out = buf.getvalue()
-        self.assertIn("Goal: build the thing", out)
+        self.assertIn("Deployed.", out)
+        self.assertIn("Status: SUCCESS", out)
+        self.assertIn("Goal: deploy the service", out)
 
     def test_final_check_mid_message_not_stripped(self):
         """Checklist bullets in the MIDDLE of a reply (followed by real
@@ -258,6 +266,15 @@ def test_stream_final_check_hidden(self):
         self.assertIn("Working...", row.plain)
         self.assertNotIn("Status:", row.plain)
 
+    def test_stream_pure_final_check_kept(self):
+        """A stream that is only the check block stays on screen — the
+        stream row must never go blank on the final reply."""
+        tui, _ = make_tui()
+        tui.stream_text = "[FINAL CHECK]\n- Status: SUCCESS"
+        row = tui._stream_row()
+        self.assertIsNotNone(row)
+        self.assertIn("Status: SUCCESS", row.plain)
+
     def test_reasoning_streams_normally(self):
         """While streaming, reasoning content shows up live like any
         other text — the collapse only happens in the final history."""