fix(retry): classify 429 as WAIT vs STOP using semantic signals

tests/providers/test_provider_error_metadata.py

@@ -26,14 +26,16 @@ def test_openai_handle_error_extracts_structured_metadata() -> None:
     err.response = _fake_response(
         status_code=409,
         headers={"retry-after-ms": "250", "x-should-retry": "false"},
-        text='{"error":"conflict"}',
+        text='{"error":{"type":"rate_limit_exceeded","code":"rate_limit_exceeded"}}',
     )
-    err.body = {"error": "conflict"}
+    err.body = {"error": {"type": "rate_limit_exceeded", "code": "rate_limit_exceeded"}}
 
     response = OpenAICompatProvider._handle_error(err)
 
     assert response.finish_reason == "error"
     assert response.error_status_code == 409
+    assert response.error_type == "rate_limit_exceeded"
+    assert response.error_code == "rate_limit_exceeded"
     assert response.error_retry_after_s == 0.25
     assert response.error_should_retry is False
 
@@ -58,11 +60,13 @@ def test_anthropic_error_response_extracts_structured_metadata() -> None:
         status_code=408,
         headers={"retry-after": "1.5", "x-should-retry": "true"},
     )
+    err.body = {"type": "error", "error": {"type": "rate_limit_error"}}
 
     response = AnthropicProvider._error_response(err)
 
     assert response.finish_reason == "error"
     assert response.error_status_code == 408
+    assert response.error_type == "rate_limit_error"
     assert response.error_retry_after_s == 1.5
     assert response.error_should_retry is True
 

tests/providers/test_provider_retry.py

@@ -297,6 +297,59 @@ async def test_chat_with_retry_retries_structured_status_code_without_keyword(mo
     assert delays == [1]
 
 
+@pytest.mark.asyncio
+async def test_chat_with_retry_stops_on_429_quota_exhausted(monkeypatch) -> None:
+    provider = ScriptedProvider([
+        LLMResponse(
+            content='{"error":{"type":"insufficient_quota","code":"insufficient_quota"}}',
+            finish_reason="error",
+            error_status_code=429,
+            error_type="insufficient_quota",
+            error_code="insufficient_quota",
+        ),
+        LLMResponse(content="ok"),
+    ])
+    delays: list[float] = []
+
+    async def _fake_sleep(delay: float) -> None:
+        delays.append(delay)
+
+    monkeypatch.setattr("nanobot.providers.base.asyncio.sleep", _fake_sleep)
+
+    response = await provider.chat_with_retry(messages=[{"role": "user", "content": "hello"}])
+
+    assert response.finish_reason == "error"
+    assert provider.calls == 1
+    assert delays == []
+
+
+@pytest.mark.asyncio
+async def test_chat_with_retry_retries_429_transient_rate_limit(monkeypatch) -> None:
+    provider = ScriptedProvider([
+        LLMResponse(
+            content='{"error":{"type":"rate_limit_exceeded","code":"rate_limit_exceeded"}}',
+            finish_reason="error",
+            error_status_code=429,
+            error_type="rate_limit_exceeded",
+            error_code="rate_limit_exceeded",
+            error_retry_after_s=0.2,
+        ),
+        LLMResponse(content="ok"),
+    ])
+    delays: list[float] = []
+
+    async def _fake_sleep(delay: float) -> None:
+        delays.append(delay)
+
+    monkeypatch.setattr("nanobot.providers.base.asyncio.sleep", _fake_sleep)
+
+    response = await provider.chat_with_retry(messages=[{"role": "user", "content": "hello"}])
+
+    assert response.content == "ok"
+    assert provider.calls == 2
+    assert delays == [0.2]
+
+
 @pytest.mark.asyncio
 async def test_chat_with_retry_retries_structured_timeout_kind(monkeypatch) -> None:
     provider = ScriptedProvider([
@@ -389,4 +442,3 @@ async def test_persistent_retry_aborts_after_ten_identical_transient_errors(monk
     assert response.content == "429 rate limit"
     assert provider.calls == 10
     assert delays == [1, 2, 4, 4, 4, 4, 4, 4, 4]
-