Python: Retry the Magentic progress ledger when it names an unknown agent

python/packages/orchestrations/agent_framework_orchestrations/_magentic.py

@@ -724,21 +724,37 @@ async def create_progress_ledger(self, magentic_context: MagenticContext) -> Mag
         # Include full context to help the model decide current stage, with small retry loop
         attempts = 0
         last_error: Exception | None = None
+        unknown_speaker_ledger: MagenticProgressLedger | None = None
         while attempts < self.progress_ledger_retry_count:
             raw = await self._complete([*magentic_context.chat_history, user_message])
             try:
                 ledger_dict = _extract_json(raw.text)
-                return _coerce_model(MagenticProgressLedger, ledger_dict)
+                ledger = _coerce_model(MagenticProgressLedger, ledger_dict)
             except Exception as ex:
                 last_error = ex
-                attempts += 1
-                logger.warning(
-                    f"Progress ledger JSON parse failed (attempt {attempts}/{self.progress_ledger_retry_count}): {ex}"
-                )
-                if attempts < self.progress_ledger_retry_count:
-                    # brief backoff before next try
-                    await asyncio.sleep(0.25 * attempts)
+                unknown_speaker_ledger = None
+            else:
+                next_speaker = ledger.next_speaker.answer
+                if (
+                    ledger.is_request_satisfied.answer
+                    or not isinstance(next_speaker, str)
+                    or next_speaker in agent_names
+                ):
+                    return ledger
+                # Retry an unknown name, but keep the ledger so running out of attempts ends the run
+                # through the orchestrator's invalid speaker handling instead of another reset.
+                unknown_speaker_ledger = ledger
+                last_error = ValueError(f"Unknown next speaker '{next_speaker}'. Valid names: {names_csv}.")
+            attempts += 1
+            logger.warning(
+                f"Progress ledger attempt failed ({attempts}/{self.progress_ledger_retry_count}): {last_error}"
+            )
+            if attempts < self.progress_ledger_retry_count:
+                # brief backoff before next try
+                await asyncio.sleep(0.25 * attempts)
 
+        if unknown_speaker_ledger is not None:
+            return unknown_speaker_ledger
         raise RuntimeError(
             f"Progress ledger parse failed after {self.progress_ledger_retry_count} attempt(s): {last_error}"
         )

python/packages/orchestrations/tests/test_magentic.py

@@ -581,6 +581,74 @@ async def fake_complete_bad(messages: list[Message], **kwargs: Any) -> Message:
         await mgr.create_progress_ledger(ctx.clone())
 
 
+async def test_standard_manager_progress_ledger_retries_unknown_next_speaker():
+    mgr = StandardMagenticManager(agent=StubManagerAgent(), progress_ledger_retry_count=2)
+    ctx = MagenticContext(task="task", participant_descriptions={"alice": "desc"})
+    speakers = ["Alice", "alice"]
+
+    async def fake_complete(messages: list[Message], **kwargs: Any) -> Message:
+        json_text = (
+            '{"is_request_satisfied": {"reason": "r", "answer": false}, '
+            '"is_in_loop": {"reason": "r", "answer": false}, '
+            '"is_progress_being_made": {"reason": "r", "answer": true}, '
+            f'"next_speaker": {{"reason": "r", "answer": "{speakers.pop(0)}"}}, '
+            '"instruction_or_question": {"reason": "r", "answer": "do"}}'
+        )
+        return Message("assistant", [json_text])
+
+    mgr._complete = fake_complete  # type: ignore[method-assign]  # ty: ignore[invalid-assignment]
+    ledger = await mgr.create_progress_ledger(ctx.clone())
+
+    assert ledger.next_speaker.answer == "alice"
+    assert not speakers
+
+
+async def test_standard_manager_progress_ledger_returns_unknown_speaker_after_retries():
+    """Exhausted retries on an unknown name end the run instead of triggering another reset."""
+    mgr = StandardMagenticManager(agent=StubManagerAgent(), progress_ledger_retry_count=3)
+    ctx = MagenticContext(task="task", participant_descriptions={"alice": "desc"})
+    calls = 0
+
+    async def fake_complete(messages: list[Message], **kwargs: Any) -> Message:
+        nonlocal calls
+        calls += 1
+        json_text = (
+            '{"is_request_satisfied": {"reason": "r", "answer": false}, '
+            '"is_in_loop": {"reason": "r", "answer": false}, '
+            '"is_progress_being_made": {"reason": "r", "answer": true}, '
+            '"next_speaker": {"reason": "r", "answer": "Alice"}, '
+            '"instruction_or_question": {"reason": "r", "answer": "do"}}'
+        )
+        return Message("assistant", [json_text])
+
+    mgr._complete = fake_complete  # type: ignore[method-assign]  # ty: ignore[invalid-assignment]
+    ledger = await mgr.create_progress_ledger(ctx.clone())
+
+    assert ledger.next_speaker.answer == "Alice"
+    assert calls == 3
+
+
+async def test_standard_manager_progress_ledger_reports_parse_failure_after_unknown_speaker():
+    """A saved unknown-speaker ledger does not hide a parse failure on the final attempt."""
+    mgr = StandardMagenticManager(agent=StubManagerAgent(), progress_ledger_retry_count=3)
+    ctx = MagenticContext(task="task", participant_descriptions={"alice": "desc"})
+    unknown_speaker = (
+        '{"is_request_satisfied": {"reason": "r", "answer": false}, '
+        '"is_in_loop": {"reason": "r", "answer": false}, '
+        '"is_progress_being_made": {"reason": "r", "answer": true}, '
+        '"next_speaker": {"reason": "r", "answer": "Alice"}, '
+        '"instruction_or_question": {"reason": "r", "answer": "do"}}'
+    )
+    replies = [unknown_speaker, "not-json", "not-json"]
+
+    async def fake_complete(messages: list[Message], **kwargs: Any) -> Message:
+        return Message("assistant", [replies.pop(0)])
+
+    mgr._complete = fake_complete  # type: ignore[method-assign]  # ty: ignore[invalid-assignment]
+    with pytest.raises(RuntimeError, match="No JSON object found"):
+        await mgr.create_progress_ledger(ctx.clone())
+
+
 class InvokeOnceManager(MagenticManagerBase):
     def __init__(self) -> None:
         super().__init__(max_round_count=5, max_stall_count=3, max_reset_count=2)