feat(mma): complete Phase 6 and finalize Comprehensive GUI UX track

- Implement Live Worker Streaming: wire ai_client.comms_log_callback to Tier 3 streams
- Add Parallel DAG Execution using asyncio.gather for non-dependent tickets
- Implement Automatic Retry with Model Escalation (Flash-Lite -> Flash -> Pro)
- Add Tier Model Configuration UI to MMA Dashboard with project TOML persistence
- Fix FPS reporting in PerformanceMonitor to prevent transient 0.0 values
- Update Ticket model with retry_count and dictionary-like access
- Stabilize Gemini CLI integration tests and handle script approval events in simulations
- Finalize and verify all 6 phases of the implementation plan

multi_agent_conductor.py

@@ -100,23 +100,35 @@ class ConductorEngine:
      print("No more executable tickets. Track is blocked or finished.")
      await self._push_state(status="blocked", active_tier=None)
     break
-    # 3. Process ready tasks
+   # 3. Process ready tasks
+   to_run = [t for t in ready_tasks if t.status == "in_progress" or (not t.step_mode and self.engine.auto_queue)]
+   
+   # Handle those awaiting approval
    for ticket in ready_tasks:
-   # If auto_queue is on and step_mode is off, engine.tick() already marked it 'in_progress'
-   # but we need to verify and handle the lifecycle.
-    if ticket.status == "in_progress" or (not ticket.step_mode and self.engine.auto_queue):
+    if ticket not in to_run and ticket.status == "todo":
+     print(f"Ticket {ticket.id} is ready and awaiting approval.")
+     await self._push_state(active_tier=f"Awaiting Approval: {ticket.id}")
+     await asyncio.sleep(1)
+
+   if to_run:
+    tasks = []
+    for ticket in to_run:
      ticket.status = "in_progress"
      print(f"Executing ticket {ticket.id}: {ticket.description}")
      await self._push_state(active_tier=f"Tier 3 (Worker): {ticket.id}")
+     
+     # Escalation logic based on retry_count
+     models = ["gemini-2.5-flash-lite", "gemini-2.5-flash", "gemini-3.1-pro-preview"]
+     model_idx = min(ticket.retry_count, len(models) - 1)
+     model_name = models[model_idx]
+
      context = WorkerContext(
       ticket_id=ticket.id,
-      model_name=self.tier_usage["Tier 3"]["model"],
+      model_name=model_name,
       messages=[]
      )
-     # Offload the blocking lifecycle call to a thread to avoid blocking the async event loop.
-     # We pass the md_content so the worker has full context.
      context_files = ticket.context_requirements if ticket.context_requirements else None
-     await loop.run_in_executor(
+     tasks.append(loop.run_in_executor(
       None,
       run_worker_lifecycle,
       ticket,
@@ -126,15 +138,19 @@ class ConductorEngine:
       self,
       md_content,
       loop
-     )
-     await self._push_state(active_tier="Tier 2 (Tech Lead)")
-    elif ticket.status == "todo" and (ticket.step_mode or not self.engine.auto_queue):
-    # Task is ready but needs approval
-     print(f"Ticket {ticket.id} is ready and awaiting approval.")
-     await self._push_state(active_tier=f"Awaiting Approval: {ticket.id}")
-     # In a real UI, this would wait for a user event.
-     # For now, we'll treat it as a pause point if not auto-queued.
-     await asyncio.sleep(1)
+     ))
+    
+    await asyncio.gather(*tasks)
+    
+    # 4. Retry and escalation logic
+    for ticket in to_run:
+     if ticket.status == 'blocked':
+      if ticket.get('retry_count', 0) < 2:
+       ticket.retry_count += 1
+       ticket.status = 'todo'
+       print(f"Ticket {ticket.id} BLOCKED. Escalating to {models[min(ticket.retry_count, len(models)-1)]} and retrying...")
+    
+    await self._push_state(active_tier="Tier 2 (Tech Lead)")
 
 def _queue_put(event_queue: events.AsyncEventQueue, loop: asyncio.AbstractEventLoop, event_name: str, payload) -> None:
  """Thread-safe helper to push an event to the AsyncEventQueue from a worker thread."""
@@ -220,6 +236,7 @@ def run_worker_lifecycle(ticket: Ticket, context: WorkerContext, context_files:
     """
  # Enforce Context Amnesia: each ticket starts with a clean slate.
  ai_client.reset_session()
+ ai_client.set_provider(ai_client.get_provider(), context.model_name)
  context_injection = ""
  if context_files:
   parser = ASTParser(language="python")
@@ -273,15 +290,37 @@ def run_worker_lifecycle(ticket: Ticket, context: WorkerContext, context_files:
   if event_queue and loop:
    _queue_put(event_queue, loop, 'mma_stream', {'stream_id': f'Tier 3 (Worker): {ticket.id}', 'text': chunk})
 
- comms_baseline = len(ai_client.get_comms_log())
- response = ai_client.send(
-  md_content=md_content,
-  user_message=user_message,
-  base_dir=".",
-  pre_tool_callback=clutch_callback if ticket.step_mode else None,
-  qa_callback=ai_client.run_tier4_analysis,
-  stream_callback=stream_callback
- )
+ old_comms_cb = ai_client.comms_log_callback
+ def worker_comms_callback(entry: dict) -> None:
+  if event_queue and loop:
+   kind = entry.get("kind")
+   payload = entry.get("payload", {})
+   chunk = ""
+   if kind == "tool_call":
+    chunk = f"\n\n[TOOL CALL] {payload.get('name')}\n{json.dumps(payload.get('script') or payload.get('args'))}\n"
+   elif kind == "tool_result":
+    res = str(payload.get("output", ""))
+    if len(res) > 500: res = res[:500] + "... (truncated)"
+    chunk = f"\n[TOOL RESULT]\n{res}\n"
+   
+   if chunk:
+    _queue_put(event_queue, loop, "response", {"text": chunk, "stream_id": f"Tier 3 (Worker): {ticket.id}", "status": "streaming..."})
+  if old_comms_cb:
+   old_comms_cb(entry)
+
+ ai_client.comms_log_callback = worker_comms_callback
+ try:
+  comms_baseline = len(ai_client.get_comms_log())
+  response = ai_client.send(
+   md_content=md_content,
+   user_message=user_message,
+   base_dir=".",
+   pre_tool_callback=clutch_callback if ticket.step_mode else None,
+   qa_callback=ai_client.run_tier4_analysis,
+   stream_callback=stream_callback
+  )
+ finally:
+  ai_client.comms_log_callback = old_comms_cb
 
  if event_queue:
   # Push via "response" event type — _process_event_queue wraps this