fix(gateway): tolerate punctuation on silence markers

gateway/response_filters.py

@@ -7,6 +7,7 @@ conversation history.
 
 from __future__ import annotations
 
+import unicodedata
 from typing import Any
 
 # Canonical model-emitted control token for intentional silence.
@@ -27,6 +28,31 @@ def _canonical_silence_candidate(text: str) -> str:
     return " ".join(text.strip().upper().split())
 
 
+def _strip_edge_silence_punctuation(text: str) -> str:
+    """Strip stray edge punctuation without erasing marker structure.
+
+    Models sometimes emit ``.NO_REPLY`` or ``*NO_REPLY*`` instead of the exact
+    marker. Keep square brackets structural so malformed ``[SILENT`` does not
+    become ``SILENT``.
+    """
+    start = 0
+    end = len(text)
+    while start < end and text[start] not in "[]" and unicodedata.category(text[start]).startswith("P"):
+        start += 1
+    while end > start and text[end - 1] not in "[]" and unicodedata.category(text[end - 1]).startswith("P"):
+        end -= 1
+    return text[start:end].strip()
+
+
+def _canonical_silence_candidates(text: str) -> tuple[str, ...]:
+    exact = _canonical_silence_candidate(text)
+    stripped = _strip_edge_silence_punctuation(text.strip())
+    if stripped == text.strip():
+        return (exact,)
+    fallback = _canonical_silence_candidate(stripped)
+    return (exact, fallback)
+
+
 def is_intentional_silence_response(response: Any) -> bool:
     """Return True only when ``response`` is exactly a silence marker.
 
@@ -41,7 +67,7 @@ def is_intentional_silence_response(response: Any) -> bool:
         return False
     if len(stripped) > 64:
         return False
-    return _canonical_silence_candidate(stripped) in LIVE_GATEWAY_SILENT_MARKERS
+    return any(candidate in LIVE_GATEWAY_SILENT_MARKERS for candidate in _canonical_silence_candidates(stripped))
 
 
 def is_intentional_silence_agent_result(agent_result: dict | None, response: Any) -> bool:
@@ -74,7 +100,7 @@ def is_partial_silence_marker(text: Any) -> bool:
     stripped = text.strip()
     if not stripped or len(stripped) > 64:
         return False
-    candidate = _canonical_silence_candidate(stripped)
-    if not candidate:
-        return False
-    return any(marker.startswith(candidate) for marker in LIVE_GATEWAY_SILENT_MARKERS)
+    for candidate in _canonical_silence_candidates(stripped):
+        if candidate and any(marker.startswith(candidate) for marker in LIVE_GATEWAY_SILENT_MARKERS):
+            return True
+    return False

tests/gateway/test_response_filters.py

@@ -9,10 +9,17 @@ def test_exact_silence_tokens_are_intentional_silence():
         assert is_intentional_silence_response(token)
 
 
+def test_edge_punctuation_silence_tokens_are_intentional_silence():
+    for token in (".NO_REPLY", "*NO_REPLY*", " .NO_REPLY ", "*[SILENT]*", "NO_REPLY."):
+        assert is_intentional_silence_response(token)
+
+
 def test_blank_and_prose_mentions_are_not_silence():
     assert not is_intentional_silence_response("")
     assert not is_intentional_silence_response("Use NO_REPLY when no answer is needed.")
     assert not is_intentional_silence_response("The reply was [SILENT], intentionally.")
+    assert not is_intentional_silence_response("😄 NO_REPLY")
+    assert not is_intentional_silence_response("[SILENT")
 
 
 def test_failed_agent_result_never_counts_as_intentional_silence():