feat: per-model max_rounds for Gemini orchestrator engine

Mirrors the pattern already in openai_orchestrator.py. The Gemini engine
was still hardcoded to the global orchestrator_max_rounds setting.

- orchestrator_engine.py: max_rounds param on run() and _run_from_contents();
  effective_limit = min(per_model_limit, global_limit); stored in checkpoint
  so resume() respects it across confirmation gates
- routers/orchestrator.py: passes orch_model.get("max_rounds") to run()
- tools/agents.py: passes model_cfg.get("max_rounds") for gemini_api spawns

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

cortex/orchestrator_engine.py

@@ -91,6 +91,7 @@ class OrchestrateCheckpoint:
     confirm_allow: frozenset = field(default_factory=frozenset)
     confirm_deny: frozenset = field(default_factory=frozenset)
     rounds_used: int = 0
+    max_rounds: int | None = None
 
 
 @dataclass
@@ -114,6 +115,7 @@ async def run(
     tool_list: list[str] | None = None,
     confirm_allow: set[str] | None = None,
     confirm_deny: set[str] | None = None,
+    max_rounds: int | None = None,
 ) -> OrchestratorResult:
     """
     Run the full orchestration loop for a task.
@@ -173,6 +175,7 @@ async def run(
         confirm_deny=_confirm_deny,
         starting_round=0,
         gemini_api_key=api_key,
+        max_rounds=max_rounds,
     )
 
     if checkpoint:
@@ -248,6 +251,7 @@ async def resume(checkpoint: OrchestrateCheckpoint, confirmed: bool) -> Orchestr
         confirm_deny=checkpoint.confirm_deny,
         starting_round=checkpoint.rounds_used,
         gemini_api_key=api_key,
+        max_rounds=checkpoint.max_rounds,
     )
 
     if new_checkpoint:
@@ -289,14 +293,17 @@ async def _run_from_contents(
     starting_round: int = 0,
     gemini_api_key: str | None = None,
     tool_list: list[str] | None = None,
+    max_rounds: int | None = None,
 ) -> tuple[str, OrchestrateCheckpoint | None]:
     """
     Run the ReAct loop from the current contents state.
     Returns (gemini_summary, checkpoint) — checkpoint is set if confirmation is needed.
     """
     gemini_summary = ""
+    per_model_limit = max_rounds or settings.orchestrator_max_rounds
+    effective_limit = min(per_model_limit, settings.orchestrator_max_rounds)
 
-    for round_num in range(starting_round, settings.orchestrator_max_rounds):
+    for round_num in range(starting_round, effective_limit):
         logger.info("Orchestrator round %d for task: %.80s", round_num + 1, task)
 
         response = await asyncio.to_thread(
@@ -400,15 +407,16 @@ async def _run_from_contents(
                 confirm_allow=confirm_allow,
                 confirm_deny=confirm_deny,
                 rounds_used=round_num + 2,
+                max_rounds=max_rounds,
             )
             return gemini_summary, checkpoint
 
         contents.append(types.Content(role="user", parts=response_parts))
 
     else:
-        logger.warning("Orchestrator hit max rounds (%d)", settings.orchestrator_max_rounds)
+        logger.warning("Orchestrator hit max rounds (%d)", effective_limit)
         gemini_summary = (
-            f"Reached the tool iteration limit ({settings.orchestrator_max_rounds} rounds). "
+            f"Reached the tool iteration limit ({effective_limit} rounds). "
             "Here is what was gathered so far:\n\n"
             + "\n\n".join(f"**{t['tool']}**: {t['result'][:500]}" for t in tool_call_log)
         )

cortex/routers/orchestrator.py

@@ -247,6 +247,7 @@ async def _run_job(job_id: str, req: OrchestrateRequest, user: str) -> None:
                 tool_list=tool_list,
                 confirm_allow=confirm_allow,
                 confirm_deny=confirm_deny,
+                max_rounds=orch_model.get("max_rounds") if orch_model else None,
             )
 
         if result.checkpoint:

cortex/tools/agents.py

@@ -132,6 +132,7 @@ async def spawn_agent(
             tool_list=tool_list,
             confirm_allow=confirm_allow,
             confirm_deny=confirm_deny,
+            max_rounds=model_cfg.get("max_rounds"),
         )
         if result.checkpoint:
             return (