feat(events): align message events with checkpoint format and add middleware tag injection

- Message events (ai_message, ai_tool_call, tool_result, human_message) now use
  BaseMessage.model_dump() format, matching LangGraph checkpoint values.messages
- on_tool_end extracts tool_call_id/name/status from ToolMessage objects
- on_tool_error now emits tool_result message events with error status
- record_middleware uses middleware:{tag} event_type and middleware category
- Summarization custom events use middleware:summarize category
- TitleMiddleware injects middleware:title tag via get_config() inheritance
- SummarizationMiddleware model bound with middleware:summarize tag
- Worker writes human_message using HumanMessage.model_dump()

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/packages/harness/deerflow/agents/lead_agent/agent.py

@@ -56,13 +56,15 @@ def _create_summarization_middleware() -> SummarizationMiddleware | None:
     # Prepare keep parameter
     keep = config.keep.to_tuple()
 
-    # Prepare model parameter
+    # Prepare model parameter.
+    # Bind "middleware:summarize" tag so RunJournal identifies these LLM calls
+    # as middleware rather than lead_agent (SummarizationMiddleware is a
+    # LangChain built-in, so we tag the model at creation time).
     if config.model_name:
         model = create_chat_model(name=config.model_name, thinking_enabled=False)
     else:
-        # Use a lightweight model for summarization to save costs
-        # Falls back to default model if not explicitly specified
         model = create_chat_model(thinking_enabled=False)
+    model = model.with_config(tags=["middleware:summarize"])
 
     # Prepare kwargs
     kwargs = {

backend/packages/harness/deerflow/agents/middlewares/title_middleware.py

@@ -1,10 +1,11 @@
 """Middleware for automatic thread title generation."""
 
 import logging
-from typing import NotRequired, override
+from typing import Any, NotRequired, override
 
 from langchain.agents import AgentState
 from langchain.agents.middleware import AgentMiddleware
+from langgraph.config import get_config
 from langgraph.runtime import Runtime
 
 from deerflow.config.title_config import get_title_config
@@ -100,6 +101,20 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
             return user_msg[:fallback_chars].rstrip() + "..."
         return user_msg if user_msg else "New Conversation"
 
+    def _get_runnable_config(self) -> dict[str, Any]:
+        """Inherit the parent RunnableConfig and add middleware tag.
+
+        This ensures RunJournal identifies LLM calls from this middleware
+        as ``middleware:title`` instead of ``lead_agent``.
+        """
+        try:
+            parent = get_config()
+        except Exception:
+            parent = {}
+        config = {**parent}
+        config["tags"] = [*(config.get("tags") or []), "middleware:title"]
+        return config
+
     def _generate_title_result(self, state: TitleMiddlewareState) -> dict | None:
         """Synchronously generate a title. Returns state update or None."""
         if not self._should_generate_title(state):
@@ -110,7 +125,7 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
         model = create_chat_model(name=config.model_name, thinking_enabled=False)
 
         try:
-            response = model.invoke(prompt)
+            response = model.invoke(prompt, config=self._get_runnable_config())
             title = self._parse_title(response.content)
             if not title:
                 title = self._fallback_title(user_msg)
@@ -130,7 +145,7 @@ class TitleMiddleware(AgentMiddleware[TitleMiddlewareState]):
         model = create_chat_model(name=config.model_name, thinking_enabled=False)
 
         try:
-            response = await model.ainvoke(prompt)
+            response = await model.ainvoke(prompt, config=self._get_runnable_config())
             title = self._parse_title(response.content)
             if not title:
                 title = self._fallback_title(user_msg)

backend/packages/harness/deerflow/runtime/journal.py

@@ -179,7 +179,8 @@ class RunJournal(BaseCallbackHandler):
             },
         )
 
-        # Message events: only lead_agent gets message-category events
+        # Message events: only lead_agent gets message-category events.
+        # Content uses message.model_dump() to align with checkpoint format.
         tool_calls = getattr(message, "tool_calls", None) or []
         if caller == "lead_agent":
             resp_meta = getattr(message, "response_metadata", None) or {}
@@ -189,7 +190,7 @@ class RunJournal(BaseCallbackHandler):
                 self._put(
                     event_type="ai_tool_call",
                     category="message",
-                    content=langchain_to_openai_message(message),
+                    content=message.model_dump(),
                     metadata={"model_name": model_name, "finish_reason": "tool_calls"},
                 )
             elif isinstance(content, str) and content:
@@ -197,10 +198,10 @@ class RunJournal(BaseCallbackHandler):
                 self._put(
                     event_type="ai_message",
                     category="message",
-                    content={"role": "assistant", "content": content},
+                    content=message.model_dump(),
                     metadata={"model_name": model_name, "finish_reason": "stop"},
                 )
-                self._last_ai_msg = content[:2000]
+                self._last_ai_msg = content
                 self._msg_count += 1
 
         # Token accumulation
@@ -242,45 +243,87 @@ class RunJournal(BaseCallbackHandler):
             },
         )
 
-    def on_tool_end(self, output: str, *, run_id: UUID, **kwargs: Any) -> None:
-        tool_call_id = kwargs.get("tool_call_id") or self._tool_call_ids.pop(str(run_id), None)
-        tool_name = kwargs.get("name", "")
+    def on_tool_end(self, output: Any, *, run_id: UUID, **kwargs: Any) -> None:
+        from langchain_core.messages import ToolMessage
+
+        # Extract fields from ToolMessage object when LangChain provides one.
+        # LangChain's _format_output wraps tool results into a ToolMessage
+        # with tool_call_id, name, status, and artifact — more complete than
+        # what kwargs alone provides.
+        if isinstance(output, ToolMessage):
+            tool_call_id = output.tool_call_id or kwargs.get("tool_call_id") or self._tool_call_ids.pop(str(run_id), None)
+            tool_name = output.name or kwargs.get("name", "")
+            status = getattr(output, "status", "success") or "success"
+            content_str = output.content if isinstance(output.content, str) else str(output.content)
+            # Use model_dump() for checkpoint-aligned message content.
+            # Override tool_call_id if it was resolved from cache.
+            msg_content = output.model_dump()
+            if msg_content.get("tool_call_id") != tool_call_id:
+                msg_content["tool_call_id"] = tool_call_id
+        else:
+            tool_call_id = kwargs.get("tool_call_id") or self._tool_call_ids.pop(str(run_id), None)
+            tool_name = kwargs.get("name", "")
+            status = "success"
+            content_str = str(output)
+            # Construct checkpoint-aligned dict when output is a plain string.
+            msg_content = ToolMessage(
+                content=content_str,
+                tool_call_id=tool_call_id or "",
+                name=tool_name,
+                status=status,
+            ).model_dump()
 
         # Trace event (always)
         self._put(
             event_type="tool_end",
             category="trace",
-            content=str(output),
+            content=content_str,
             metadata={
                 "tool_name": tool_name,
                 "tool_call_id": tool_call_id,
-                "status": "success",
+                "status": status,
             },
         )
 
-        # Message event: tool_result
+        # Message event: tool_result (checkpoint-aligned model_dump format)
         self._put(
             event_type="tool_result",
             category="message",
-            content={
-                "role": "tool",
-                "tool_call_id": tool_call_id or "",
-                "content": str(output),
-            },
-            metadata={"tool_name": tool_name},
+            content=msg_content,
+            metadata={"tool_name": tool_name, "status": status},
         )
 
     def on_tool_error(self, error: BaseException, *, run_id: UUID, **kwargs: Any) -> None:
+        from langchain_core.messages import ToolMessage
+
+        tool_call_id = kwargs.get("tool_call_id") or self._tool_call_ids.pop(str(run_id), None)
+        tool_name = kwargs.get("name", "")
+
+        # Trace event
         self._put(
             event_type="tool_error",
             category="trace",
             content=str(error),
             metadata={
-                "tool_name": kwargs.get("name", ""),
-                "tool_call_id": kwargs.get("tool_call_id"),
+                "tool_name": tool_name,
+                "tool_call_id": tool_call_id,
             },
         )
 
+        # Message event: tool_result with error status (checkpoint-aligned)
+        msg_content = ToolMessage(
+            content=str(error),
+            tool_call_id=tool_call_id or "",
+            name=tool_name,
+            status="error",
+        ).model_dump()
+        self._put(
+            event_type="tool_result",
+            category="message",
+            content=msg_content,
+            metadata={"tool_name": tool_name, "status": "error"},
+        )
+
     # -- Custom event callback --
 
     def on_custom_event(self, name: str, data: Any, *, run_id: UUID, **kwargs: Any) -> None:
@@ -298,8 +341,8 @@ class RunJournal(BaseCallbackHandler):
                 },
             )
             self._put(
-                event_type="summary",
-                category="message",
+                event_type="middleware:summarize",
+                category="middleware",
                 content={"role": "system", "content": data_dict.get("summary", "")},
                 metadata={"replaced_count": data_dict.get("replaced_count", 0)},
             )
@@ -366,16 +409,24 @@ class RunJournal(BaseCallbackHandler):
         """Record the first human message for convenience fields."""
         self._first_human_msg = content[:2000] if content else None
 
-    def record_middleware(self, name: str, hook: str, action: str, changes: dict) -> None:
-        """Record a middleware trace event.
+    def record_middleware(self, tag: str, *, name: str, hook: str, action: str, changes: dict) -> None:
+        """Record a middleware state-change event.
 
         Called by middleware implementations when they perform a meaningful
         state change (e.g., title generation, summarization, HITL approval).
         Pure-observation middleware should not call this.
+
+        Args:
+            tag: Short identifier for the middleware (e.g., "title", "summarize",
+                 "guardrail"). Used to form event_type="middleware:{tag}".
+            name: Full middleware class name.
+            hook: Lifecycle hook that triggered the action (e.g., "after_model").
+            action: Specific action performed (e.g., "generate_title").
+            changes: Dict describing the state changes made.
         """
         self._put(
-            event_type="middleware",
-            category="trace",
+            event_type=f"middleware:{tag}",
+            category="middleware",
             content={"name": name, "hook": hook, "action": action, "changes": changes},
         )
 

backend/packages/harness/deerflow/runtime/runs/worker.py

@@ -67,9 +67,9 @@ async def run_agent(
             track_token_usage=getattr(run_events_config, "track_token_usage", True),
         )
 
-        # Write human_message event
-        user_input = _extract_user_input(graph_input)
-        if user_input:
+        # Write human_message event (model_dump format, aligned with checkpoint)
+        human_msg = _extract_human_message(graph_input)
+        if human_msg is not None:
             msg_metadata = {}
             if follow_up_to_run_id:
                 msg_metadata["follow_up_to_run_id"] = follow_up_to_run_id
@@ -78,10 +78,11 @@ async def run_agent(
                 run_id=run_id,
                 event_type="human_message",
                 category="message",
-                content={"role": "user", "content": user_input},
+                content=human_msg.model_dump(),
                 metadata=msg_metadata or None,
             )
-            journal.set_first_human_message(user_input)
+            content = human_msg.content
+            journal.set_first_human_message(content if isinstance(content, str) else str(content))
 
     # Track whether "events" was requested but skipped
     if "events" in requested_modes:
@@ -282,21 +283,29 @@ def _lg_mode_to_sse_event(mode: str) -> str:
     return mode
 
 
-def _extract_user_input(graph_input: dict) -> str:
-    """Extract user input text from graph_input for event recording."""
+def _extract_human_message(graph_input: dict) -> "HumanMessage | None":
+    """Extract or construct a HumanMessage from graph_input for event recording.
+
+    Returns a LangChain HumanMessage so callers can use .model_dump() to get
+    the checkpoint-aligned serialization format.
+    """
+    from langchain_core.messages import HumanMessage
+
     messages = graph_input.get("messages")
     if not messages:
-        return ""
-    # Take the last message (usually the user's input)
+        return None
     last = messages[-1] if isinstance(messages, list) else messages
-    if isinstance(last, str):
+    if isinstance(last, HumanMessage):
         return last
+    if isinstance(last, str):
+        return HumanMessage(content=last) if last else None
     if hasattr(last, "content"):
         content = last.content
-        return content if isinstance(content, str) else str(content)
+        return HumanMessage(content=content)
     if isinstance(last, dict):
-        return str(last.get("content", ""))
-    return ""
+        content = last.get("content", "")
+        return HumanMessage(content=content) if content else None
+    return None
 
 
 def _unpack_stream_item(

backend/tests/test_lead_agent_model_resolution.py

@@ -146,8 +146,11 @@ def test_create_summarization_middleware_uses_configured_model_alias(monkeypatch
         lambda: SummarizationConfig(enabled=True, model_name="model-masswork"),
     )
 
+    from unittest.mock import MagicMock
+
     captured: dict[str, object] = {}
-    fake_model = object()
+    fake_model = MagicMock()
+    fake_model.with_config.return_value = fake_model
 
     def _fake_create_chat_model(*, name=None, thinking_enabled, reasoning_effort=None):
         captured["name"] = name
@@ -163,3 +166,4 @@ def test_create_summarization_middleware_uses_configured_model_alias(monkeypatch
     assert captured["name"] == "model-masswork"
     assert captured["thinking_enabled"] is False
     assert middleware["model"] is fake_model
+    fake_model.with_config.assert_called_once_with(tags=["middleware:summarize"])

backend/tests/test_run_journal.py

@@ -20,24 +20,32 @@ def journal_setup():
     return j, store
 
 
-def _make_llm_response(content="Hello", usage=None, tool_calls=None):
-    """Create a mock LLM response with a message."""
+def _make_llm_response(content="Hello", usage=None, tool_calls=None, additional_kwargs=None):
+    """Create a mock LLM response with a message.
+
+    model_dump() returns checkpoint-aligned format matching real AIMessage.
+    """
     msg = MagicMock()
     msg.type = "ai"
     msg.content = content
     msg.id = f"msg-{id(msg)}"
     msg.tool_calls = tool_calls or []
+    msg.invalid_tool_calls = []
     msg.response_metadata = {"model_name": "test-model"}
     msg.usage_metadata = usage
-    # Provide a real model_dump so serialize_lc_object returns a plain dict
-    # (needed for DB-backed tests where json.dumps must succeed).
+    msg.additional_kwargs = additional_kwargs or {}
+    msg.name = None
+    # model_dump returns checkpoint-aligned format
     msg.model_dump.return_value = {
-        "type": "ai",
         "content": content,
+        "additional_kwargs": additional_kwargs or {},
+        "response_metadata": {"model_name": "test-model"},
+        "type": "ai",
+        "name": None,
         "id": msg.id,
         "tool_calls": tool_calls or [],
+        "invalid_tool_calls": [],
         "usage_metadata": usage,
-        "response_metadata": {"model_name": "test-model"},
     }
 
     gen = MagicMock()
@@ -71,7 +79,9 @@ class TestLlmCallbacks:
         messages = await store.list_messages("t1")
         assert len(messages) == 1
         assert messages[0]["event_type"] == "ai_message"
-        assert messages[0]["content"] == {"role": "assistant", "content": "Answer"}
+        # Content is checkpoint-aligned model_dump format
+        assert messages[0]["content"]["type"] == "ai"
+        assert messages[0]["content"]["content"] == "Answer"
 
     @pytest.mark.anyio
     async def test_on_llm_end_with_tool_calls_produces_ai_tool_call(self, journal_setup):
@@ -211,10 +221,14 @@ class TestCustomEvents:
         events = await store.list_events("t1", "r1")
         trace = [e for e in events if e["event_type"] == "summarization"]
         assert len(trace) == 1
+        # Summarization goes to middleware category, not message
+        mw_events = [e for e in events if e["event_type"] == "middleware:summarize"]
+        assert len(mw_events) == 1
+        assert mw_events[0]["category"] == "middleware"
+        assert mw_events[0]["content"] == {"role": "system", "content": "Context was summarized."}
+        # No message events from summarization
         messages = await store.list_messages("t1")
-        assert len(messages) == 1
-        assert messages[0]["event_type"] == "summary"
-        assert messages[0]["content"] == {"role": "system", "content": "Context was summarized."}
+        assert len(messages) == 0
 
     @pytest.mark.anyio
     async def test_non_summarization_custom_event(self, journal_setup):
@@ -375,8 +389,11 @@ class TestDbBackedLifecycle:
         record = await mgr.create("t1", "lead_agent")
         run_id = record.run_id
 
-        # Write human_message
-        await event_store.put(thread_id="t1", run_id=run_id, event_type="human_message", category="message", content={"role": "user", "content": "Hello DB"})
+        # Write human_message (checkpoint-aligned format)
+        from langchain_core.messages import HumanMessage
+
+        human_msg = HumanMessage(content="Hello DB")
+        await event_store.put(thread_id="t1", run_id=run_id, event_type="human_message", category="message", content=human_msg.model_dump())
 
         # Simulate journal
         journal = RunJournal(run_id, "t1", event_store, flush_threshold=100)
@@ -406,12 +423,14 @@ class TestDbBackedLifecycle:
         assert row["status"] == "success"
         assert row["total_tokens"] == 15
 
-        # Verify messages from DB
+        # Verify messages from DB (checkpoint-aligned format)
         messages = await event_store.list_messages("t1")
         assert len(messages) == 2
         assert messages[0]["event_type"] == "human_message"
+        assert messages[0]["content"]["type"] == "human"
         assert messages[1]["event_type"] == "ai_message"
-        assert messages[1]["content"] == {"role": "assistant", "content": "DB response"}
+        assert messages[1]["content"]["type"] == "ai"
+        assert messages[1]["content"]["content"] == "DB response"
 
         # Verify events from DB
         events = await event_store.list_events("t1", run_id)
@@ -560,38 +579,45 @@ class TestDictContent:
         await close_engine()
 
 
-class TestOpenAIHumanMessage:
+class TestCheckpointAlignedHumanMessage:
     @pytest.mark.anyio
-    async def test_human_message_openai_format(self):
+    async def test_human_message_checkpoint_format(self):
+        """human_message content uses model_dump() checkpoint format."""
+        from langchain_core.messages import HumanMessage
+
         store = MemoryRunEventStore()
+        human_msg = HumanMessage(content="What is AI?")
         await store.put(
             thread_id="t1",
             run_id="r1",
             event_type="human_message",
             category="message",
-            content={"role": "user", "content": "What is AI?"},
+            content=human_msg.model_dump(),
             metadata={"message_id": "msg_001"},
         )
         messages = await store.list_messages("t1")
         assert len(messages) == 1
-        assert messages[0]["content"] == {"role": "user", "content": "What is AI?"}
-        assert messages[0]["content"]["role"] == "user"
+        assert messages[0]["content"]["type"] == "human"
+        assert messages[0]["content"]["content"] == "What is AI?"
 
 
-class TestOpenAIMessageFormat:
+class TestCheckpointAlignedMessageFormat:
     @pytest.mark.anyio
-    async def test_ai_message_openai_format(self, journal_setup):
-        """ai_message content should be OpenAI assistant message dict."""
+    async def test_ai_message_checkpoint_format(self, journal_setup):
+        """ai_message content should be checkpoint-aligned model_dump dict."""
         j, store = journal_setup
         j.on_llm_end(_make_llm_response("Answer"), run_id=uuid4(), tags=["lead_agent"])
         await j.flush()
         messages = await store.list_messages("t1")
         assert len(messages) == 1
-        assert messages[0]["content"] == {"role": "assistant", "content": "Answer"}
+        assert messages[0]["content"]["type"] == "ai"
+        assert messages[0]["content"]["content"] == "Answer"
+        assert "response_metadata" in messages[0]["content"]
+        assert "additional_kwargs" in messages[0]["content"]
 
     @pytest.mark.anyio
     async def test_ai_tool_call_event(self, journal_setup):
-        """LLM response with tool_calls should produce ai_tool_call message event."""
+        """LLM response with tool_calls should produce ai_tool_call with model_dump content."""
         j, store = journal_setup
         tool_calls = [{"id": "call_1", "name": "search", "args": {"query": "test"}}]
         j.on_llm_end(
@@ -603,13 +629,12 @@ class TestOpenAIMessageFormat:
         messages = await store.list_messages("t1")
         assert len(messages) == 1
         assert messages[0]["event_type"] == "ai_tool_call"
-        assert messages[0]["content"]["role"] == "assistant"
+        assert messages[0]["content"]["type"] == "ai"
         assert messages[0]["content"]["content"] == "Let me search"
         assert len(messages[0]["content"]["tool_calls"]) == 1
         tc = messages[0]["content"]["tool_calls"][0]
         assert tc["id"] == "call_1"
-        assert tc["type"] == "function"
-        assert tc["function"]["name"] == "search"
+        assert tc["name"] == "search"
 
     @pytest.mark.anyio
     async def test_ai_tool_call_only_from_lead_agent(self, journal_setup):
@@ -637,11 +662,11 @@ class TestToolResultMessage:
         messages = await store.list_messages("t1")
         assert len(messages) == 1
         assert messages[0]["event_type"] == "tool_result"
-        assert messages[0]["content"] == {
-            "role": "tool",
-            "tool_call_id": "call_abc",
-            "content": "search results here",
-        }
+        # Content is checkpoint-aligned model_dump format
+        assert messages[0]["content"]["type"] == "tool"
+        assert messages[0]["content"]["tool_call_id"] == "call_abc"
+        assert messages[0]["content"]["content"] == "search results here"
+        assert messages[0]["content"]["name"] == "web_search"
 
     @pytest.mark.anyio
     async def test_tool_result_missing_tool_call_id(self, journal_setup):
@@ -652,15 +677,128 @@ class TestToolResultMessage:
         await j.flush()
         messages = await store.list_messages("t1")
         assert len(messages) == 1
-        assert messages[0]["content"]["role"] == "tool"
+        assert messages[0]["content"]["type"] == "tool"
 
     @pytest.mark.anyio
-    async def test_tool_error_no_tool_result_message(self, journal_setup):
+    async def test_tool_end_extracts_from_tool_message_object(self, journal_setup):
+        """When LangChain passes a ToolMessage object as output, extract fields from it."""
+        from langchain_core.messages import ToolMessage
+
+        j, store = journal_setup
+        run_id = uuid4()
+        tool_msg = ToolMessage(
+            content="search results",
+            tool_call_id="call_from_obj",
+            name="web_search",
+            status="success",
+        )
+        j.on_tool_end(tool_msg, run_id=run_id)
+        await j.flush()
+
+        messages = await store.list_messages("t1")
+        assert len(messages) == 1
+        assert messages[0]["content"]["type"] == "tool"
+        assert messages[0]["content"]["tool_call_id"] == "call_from_obj"
+        assert messages[0]["content"]["content"] == "search results"
+        assert messages[0]["content"]["name"] == "web_search"
+        assert messages[0]["metadata"]["tool_name"] == "web_search"
+        assert messages[0]["metadata"]["status"] == "success"
+
+        events = await store.list_events("t1", "r1")
+        tool_end = [e for e in events if e["event_type"] == "tool_end"][0]
+        assert tool_end["metadata"]["tool_call_id"] == "call_from_obj"
+        assert tool_end["metadata"]["tool_name"] == "web_search"
+
+    @pytest.mark.anyio
+    async def test_tool_message_object_overrides_kwargs(self, journal_setup):
+        """ToolMessage object fields take priority over kwargs."""
+        from langchain_core.messages import ToolMessage
+
+        j, store = journal_setup
+        run_id = uuid4()
+        tool_msg = ToolMessage(
+            content="result",
+            tool_call_id="call_obj",
+            name="tool_a",
+            status="success",
+        )
+        # Pass different values in kwargs — ToolMessage should win
+        j.on_tool_end(tool_msg, run_id=run_id, name="tool_b", tool_call_id="call_kwarg")
+        await j.flush()
+
+        messages = await store.list_messages("t1")
+        assert messages[0]["content"]["tool_call_id"] == "call_obj"
+        assert messages[0]["content"]["name"] == "tool_a"
+        assert messages[0]["metadata"]["tool_name"] == "tool_a"
+
+    @pytest.mark.anyio
+    async def test_tool_message_error_status(self, journal_setup):
+        """ToolMessage with status='error' propagates status to metadata."""
+        from langchain_core.messages import ToolMessage
+
+        j, store = journal_setup
+        run_id = uuid4()
+        tool_msg = ToolMessage(
+            content="something went wrong",
+            tool_call_id="call_err",
+            name="web_fetch",
+            status="error",
+        )
+        j.on_tool_end(tool_msg, run_id=run_id)
+        await j.flush()
+
+        events = await store.list_events("t1", "r1")
+        tool_end = [e for e in events if e["event_type"] == "tool_end"][0]
+        assert tool_end["metadata"]["status"] == "error"
+
+        messages = await store.list_messages("t1")
+        assert messages[0]["content"]["status"] == "error"
+        assert messages[0]["metadata"]["status"] == "error"
+
+    @pytest.mark.anyio
+    async def test_tool_message_fallback_to_cache(self, journal_setup):
+        """If ToolMessage has empty tool_call_id, fall back to cache from on_tool_start."""
+        from langchain_core.messages import ToolMessage
+
+        j, store = journal_setup
+        run_id = uuid4()
+        j.on_tool_start({"name": "bash"}, "ls", run_id=run_id, tool_call_id="call_cached")
+        tool_msg = ToolMessage(
+            content="file list",
+            tool_call_id="",
+            name="bash",
+        )
+        j.on_tool_end(tool_msg, run_id=run_id)
+        await j.flush()
+
+        messages = await store.list_messages("t1")
+        assert messages[0]["content"]["tool_call_id"] == "call_cached"
+
+    @pytest.mark.anyio
+    async def test_tool_error_produces_tool_result_message(self, journal_setup):
         j, store = journal_setup
         j.on_tool_error(TimeoutError("timeout"), run_id=uuid4(), name="web_fetch", tool_call_id="call_1")
         await j.flush()
         messages = await store.list_messages("t1")
-        assert len(messages) == 0
+        assert len(messages) == 1
+        assert messages[0]["event_type"] == "tool_result"
+        assert messages[0]["content"]["type"] == "tool"
+        assert messages[0]["content"]["tool_call_id"] == "call_1"
+        assert "timeout" in messages[0]["content"]["content"]
+        assert messages[0]["content"]["status"] == "error"
+        assert messages[0]["metadata"]["status"] == "error"
+
+    @pytest.mark.anyio
+    async def test_tool_error_uses_cached_tool_call_id(self, journal_setup):
+        """on_tool_error should fall back to cached tool_call_id from on_tool_start."""
+        j, store = journal_setup
+        run_id = uuid4()
+        j.on_tool_start({"name": "web_fetch"}, "url", run_id=run_id, tool_call_id="call_cached")
+        j.on_tool_error(TimeoutError("timeout"), run_id=run_id, name="web_fetch")
+        await j.flush()
+        messages = await store.list_messages("t1")
+        assert len(messages) == 1
+        assert messages[0]["content"]["tool_call_id"] == "call_cached"
 
 
 def _make_base_messages():
@@ -745,11 +883,12 @@ class TestLlmRequestResponse:
         assert not any(e["event_type"] == "llm_start" for e in events)
 
 
-class TestMiddlewareTrace:
+class TestMiddlewareEvents:
     @pytest.mark.anyio
-    async def test_record_middleware(self, journal_setup):
+    async def test_record_middleware_uses_middleware_category(self, journal_setup):
         j, store = journal_setup
         j.record_middleware(
+            "title",
             name="TitleMiddleware",
             hook="after_model",
             action="generate_title",
@@ -757,27 +896,60 @@ class TestMiddlewareTrace:
         )
         await j.flush()
         events = await store.list_events("t1", "r1")
-        mw_events = [e for e in events if e["event_type"] == "middleware"]
+        mw_events = [e for e in events if e["event_type"] == "middleware:title"]
         assert len(mw_events) == 1
-        assert mw_events[0]["category"] == "trace"
+        assert mw_events[0]["category"] == "middleware"
         assert mw_events[0]["content"]["name"] == "TitleMiddleware"
         assert mw_events[0]["content"]["hook"] == "after_model"
         assert mw_events[0]["content"]["action"] == "generate_title"
         assert mw_events[0]["content"]["changes"]["title"] == "Test Title"
 
+    @pytest.mark.anyio
+    async def test_middleware_events_not_in_messages(self, journal_setup):
+        """Middleware events should not appear in list_messages()."""
+        j, store = journal_setup
+        j.record_middleware(
+            "title",
+            name="TitleMiddleware",
+            hook="after_model",
+            action="generate_title",
+            changes={"title": "Test"},
+        )
+        await j.flush()
+        messages = await store.list_messages("t1")
+        assert len(messages) == 0
+
+    @pytest.mark.anyio
+    async def test_middleware_tag_variants(self, journal_setup):
+        """Different middleware tags produce distinct event_types."""
+        j, store = journal_setup
+        j.record_middleware("title", name="TitleMiddleware", hook="after_model", action="generate_title", changes={})
+        j.record_middleware("guardrail", name="GuardrailMiddleware", hook="before_tool", action="deny", changes={})
+        await j.flush()
+        events = await store.list_events("t1", "r1")
+        event_types = {e["event_type"] for e in events}
+        assert "middleware:title" in event_types
+        assert "middleware:guardrail" in event_types
+
 
 class TestFullRunSequence:
     @pytest.mark.anyio
     async def test_complete_run_event_sequence(self):
-        """Simulate a full run: user -> LLM -> tool_call -> tool_result -> LLM -> final reply."""
+        """Simulate a full run: user -> LLM -> tool_call -> tool_result -> LLM -> final reply.
+
+        All message events use checkpoint-aligned model_dump format.
+        """
+        from langchain_core.messages import HumanMessage
+
         store = MemoryRunEventStore()
         j = RunJournal("r1", "t1", store, flush_threshold=100)
 
-        # 1. Human message (written by worker, not journal)
+        # 1. Human message (written by worker, using model_dump format)
+        human_msg = HumanMessage(content="Search for quantum computing")
         await store.put(
             thread_id="t1", run_id="r1",
             event_type="human_message", category="message",
-            content={"role": "user", "content": "Search for quantum computing"},
+            content=human_msg.model_dump(),
         )
         j.set_first_human_message("Search for quantum computing")
 
@@ -805,7 +977,7 @@ class TestFullRunSequence:
         j.on_tool_end("Quantum computing results...", run_id=tool_id, name="web_search", tool_call_id="call_1")
 
         # 5. Middleware: title generation
-        j.record_middleware("TitleMiddleware", "after_model", "generate_title", {"title": "Quantum Computing"})
+        j.record_middleware("title", name="TitleMiddleware", hook="after_model", action="generate_title", changes={"title": "Quantum Computing"})
 
         # 6. Second LLM call -> final reply
         llm2_id = uuid4()
@@ -824,18 +996,19 @@ class TestFullRunSequence:
         await asyncio.sleep(0.05)
         await j.flush()
 
-        # Verify message sequence (what gets exported for training)
+        # Verify message sequence
         messages = await store.list_messages("t1")
         msg_types = [m["event_type"] for m in messages]
         assert msg_types == ["human_message", "ai_tool_call", "tool_result", "ai_message"]
 
-        # Verify message content format
-        assert messages[0]["content"]["role"] == "user"
-        assert messages[1]["content"]["role"] == "assistant"
+        # Verify checkpoint-aligned format: all messages use "type" not "role"
+        assert messages[0]["content"]["type"] == "human"
+        assert messages[0]["content"]["content"] == "Search for quantum computing"
+        assert messages[1]["content"]["type"] == "ai"
         assert "tool_calls" in messages[1]["content"]
-        assert messages[2]["content"]["role"] == "tool"
+        assert messages[2]["content"]["type"] == "tool"
         assert messages[2]["content"]["tool_call_id"] == "call_1"
-        assert messages[3]["content"]["role"] == "assistant"
+        assert messages[3]["content"]["type"] == "ai"
         assert messages[3]["content"]["content"] == "Here are the results about quantum computing..."
 
         # Verify trace events
@@ -845,10 +1018,14 @@ class TestFullRunSequence:
         assert "llm_response" in trace_types
         assert "tool_start" in trace_types
         assert "tool_end" in trace_types
-        assert "middleware" in trace_types
         assert "llm_start" not in trace_types
         assert "llm_end" not in trace_types
 
+        # Verify middleware events are in their own category
+        mw_events = [e for e in events if e["category"] == "middleware"]
+        assert len(mw_events) == 1
+        assert mw_events[0]["event_type"] == "middleware:title"
+
         # Verify token accumulation
         data = j.get_completion_data()
         assert data["total_tokens"] == 420  # 120 + 300
@@ -857,7 +1034,7 @@ class TestFullRunSequence:
         assert data["message_count"] == 1  # only final ai_message counts
         assert data["last_ai_message"] == "Here are the results about quantum computing..."
 
-        # Verify training data export is trivial
-        training_messages = [m["content"] for m in messages]
-        assert all(isinstance(m, dict) for m in training_messages)
-        assert all("role" in m for m in training_messages)
+        # Verify all message contents are checkpoint-aligned dicts with "type" field
+        for m in messages:
+            assert isinstance(m["content"], dict)
+            assert "type" in m["content"]