fix(api-server): collapse tool start/lifecycle into a single SSE event

Address Copilot review on PR #16666:

1. **Duplicate event on every tool start** — both ``tool_progress_callback``
   and ``tool_start_callback`` fire side-by-side in ``run_agent.py``, so
   wiring both into chat completions emitted *two* ``hermes.tool.progress``
   events per real tool call. Drop the legacy ``_on_tool_progress`` emit
   entirely; ``_on_tool_start`` now produces a single unified event that
   carries the legacy ``tool``/``emoji``/``label`` fields plus the new
   ``toolCallId``/``status`` correlation fields. Label is computed inline
   via ``build_tool_preview`` so callers do not need to pre-format it.

2. **Weak per-event correlation in the regression test** — the previous
   assertion checked that a ``toolCallId`` appeared *somewhere* in the
   aggregate, which would have passed even if ``running`` lacked the id.
   Collect ``(status, toolCallId)`` per event and assert each event
   carries the correct pair, plus exactly two events on the wire (no
   silent duplication regression).

The two existing chat-completions tool-progress tests are updated to fire
``tool_start_callback`` instead of ``tool_progress_callback``, matching
production reality where ``run_agent`` always pairs them.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

gateway/platforms/api_server.py

@@ -981,39 +981,62 @@ class APIServerAdapter(BasePlatformAdapter):
                 if delta is not None:
                     _stream_q.put(delta)
 
-            def _on_tool_progress(event_type, name, preview, args, **kwargs):
-                """Send tool progress as a separate SSE event.
+            # Track which tool_call_ids we've emitted a "running" lifecycle
+            # event for, so a "completed" event without a matching "running"
+            # (e.g. internal/filtered tools) is silently dropped instead of
+            # producing an orphaned event clients can't correlate.
+            _started_tool_call_ids: set[str] = set()
 
-                Previously, progress markers like ``⏰ list`` were injected
-                directly into ``delta.content``.  OpenAI-compatible frontends
-                (Open WebUI, LobeChat, …) store ``delta.content`` verbatim as
-                the assistant message and send it back on subsequent requests.
-                After enough turns the model learns to *emit* the markers as
-                plain text instead of issuing real tool calls — silently
-                hallucinating tool results.  See #6972.
+            def _on_tool_start(tool_call_id, function_name, function_args):
+                """Emit ``hermes.tool.progress`` with ``status: running``.
 
-                The fix: push a tagged tuple ``("__tool_progress__", payload)``
-                onto the stream queue.  The SSE writer emits it as a custom
-                ``event: hermes.tool.progress`` line that compliant frontends
-                can render for UX but will *not* persist into conversation
-                history.  Clients that don't understand the custom event type
-                silently ignore it per the SSE specification.
+                Replaces the old ``tool_progress_callback("tool.started",
+                ...)`` emit so SSE consumers receive a single event per
+                tool start, carrying both the legacy ``tool``/``emoji``/
+                ``label`` payload (for #6972 frontends) and the new
+                ``toolCallId``/``status`` correlation fields (#16588).
+
+                Skips tools whose names start with ``_`` so internal
+                events (``_thinking``, …) stay off the wire — matching
+                the prior ``_on_tool_progress`` filter exactly.
                 """
-                if event_type != "tool.started":
+                if not tool_call_id or function_name.startswith("_"):
                     return
-                if name.startswith("_"):
-                    return
-                from agent.display import get_tool_emoji
-                emoji = get_tool_emoji(name)
-                label = preview or name
+                _started_tool_call_ids.add(tool_call_id)
+                from agent.display import build_tool_preview, get_tool_emoji
+                label = build_tool_preview(function_name, function_args) or function_name
                 _stream_q.put(("__tool_progress__", {
-                    "tool": name,
-                    "emoji": emoji,
+                    "tool": function_name,
+                    "emoji": get_tool_emoji(function_name),
                     "label": label,
+                    "toolCallId": tool_call_id,
+                    "status": "running",
+                }))
+
+            def _on_tool_complete(tool_call_id, function_name, function_args, function_result):
+                """Emit the matching ``status: completed`` event.
+
+                Dropped if the start was filtered (internal tool, missing
+                id, or never seen) so clients never get an orphaned
+                ``completed`` they can't correlate to a prior ``running``.
+                """
+                if not tool_call_id or tool_call_id not in _started_tool_call_ids:
+                    return
+                _started_tool_call_ids.discard(tool_call_id)
+                _stream_q.put(("__tool_progress__", {
+                    "tool": function_name,
+                    "toolCallId": tool_call_id,
+                    "status": "completed",
                 }))
 
             # Start agent in background.  agent_ref is a mutable container
             # so the SSE writer can interrupt the agent on client disconnect.
+            #
+            # ``tool_progress_callback`` is intentionally not wired here:
+            # it would duplicate every emit because ``run_agent`` fires it
+            # side-by-side with ``tool_start_callback``/``tool_complete_callback``.
+            # The structured callbacks are strictly richer (they carry the
+            # tool_call id), so they own the chat-completions SSE channel.
             agent_ref = [None]
             agent_task = asyncio.ensure_future(self._run_agent(
                 user_message=user_message,
@@ -1021,7 +1044,8 @@ class APIServerAdapter(BasePlatformAdapter):
                 ephemeral_system_prompt=system_prompt,
                 session_id=session_id,
                 stream_delta_callback=_on_delta,
-                tool_progress_callback=_on_tool_progress,
+                tool_start_callback=_on_tool_start,
+                tool_complete_callback=_on_tool_complete,
                 agent_ref=agent_ref,
             ))
 
@@ -1136,7 +1160,8 @@ class APIServerAdapter(BasePlatformAdapter):
                 Tagged tuples ``("__tool_progress__", payload)`` are sent
                 as a custom ``event: hermes.tool.progress`` SSE event so
                 frontends can display them without storing the markers in
-                conversation history.  See #6972.
+                conversation history.  See #6972 for the original event,
+                #16588 for the ``toolCallId``/``status`` lifecycle fields.
                 """
                 if isinstance(item, tuple) and len(item) == 2 and item[0] == "__tool_progress__":
                     event_data = json.dumps(item[1])