From 799ef24842433de1c0225e7bf112ebcf5eb4ca40 Mon Sep 17 00:00:00 2001
From: Ousama Ben Younes <benyounes.ousama@gmail.com>
Date: Mon, 27 Jul 2026 19:45:51 +0000
Subject: [PATCH] fix(runtime): make child agent limit reservation atomic

---
 strix/core/agents.py    | 28 +++++++++++++++++++++++
 strix/core/execution.py | 24 +++++++++-----------
 tests/test_execution.py | 49 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 87 insertions(+), 14 deletions(-)

diff --git a/strix/core/agents.py b/strix/core/agents.py
index c96204df..d72236e4 100644
--- a/strix/core/agents.py
+++ b/strix/core/agents.py
@@ -168,6 +168,34 @@ class AgentCoordinator:
         logger.info("agent.register %s (%s) parent=%s", agent_id, name, parent_id or "-")
         await self._maybe_snapshot()
 
+    async def register_child_if_capacity(
+        self,
+        agent_id: str,
+        name: str,
+        parent_id: str,
+        *,
+        max_child_agents: int,
+        task: str | None = None,
+        skills: list[str] | None = None,
+    ) -> tuple[bool, int]:
+        async with self._lock:
+            child_count = sum(parent is not None for parent in self.parent_of.values())
+            if max_child_agents > 0 and child_count >= max_child_agents:
+                return False, child_count
+            self.statuses[agent_id] = "running"
+            self.parent_of[agent_id] = parent_id
+            self.names[agent_id] = name
+            self.pending_counts.setdefault(agent_id, 0)
+            self.metadata[agent_id] = {
+                "task": task or "",
+                "skills": list(skills or []),
+            }
+            self.runtimes.setdefault(agent_id, AgentRuntime())
+            child_count += 1
+        logger.info("agent.register %s (%s) parent=%s", agent_id, name, parent_id)
+        await self._maybe_snapshot()
+        return True, child_count
+
     async def attach_runtime(
         self,
         agent_id: str,
diff --git a/strix/core/execution.py b/strix/core/execution.py
index 6b7c3462..b8ce1cfa 100644
--- a/strix/core/execution.py
+++ b/strix/core/execution.py
@@ -310,8 +310,16 @@ async def spawn_child_agent(
     if not isinstance(parent_id, str):
         raise TypeError("Parent agent_id missing from context")
 
-    child_count = _child_agent_count(coordinator)
-    if max_child_agents > _UNLIMITED_CHILD_AGENTS and child_count >= max_child_agents:
+    child_id = uuid.uuid4().hex[:8]
+    registered, child_count = await coordinator.register_child_if_capacity(
+        child_id,
+        name,
+        parent_id,
+        max_child_agents=max_child_agents,
+        task=task,
+        skills=skills,
+    )
+    if not registered:
         logger.info(
             "refusing to spawn child agent %r: limit %d already reached",
             name,
@@ -328,15 +336,7 @@ async def spawn_child_agent(
             "current_child_agents": child_count,
         }
 
-    child_id = uuid.uuid4().hex[:8]
     child_agent = factory(name=name, skills=skills)
-    await coordinator.register(
-        child_id,
-        name,
-        parent_id,
-        task=task,
-        skills=skills,
-    )
 
     await _start_child_runner(
         parent_ctx=parent_ctx,
@@ -371,10 +371,6 @@ async def spawn_child_agent(
     }
 
 
-def _child_agent_count(coordinator: AgentCoordinator) -> int:
-    return sum(parent_id is not None for parent_id in coordinator.parent_of.values())
-
-
 async def respawn_subagents(
     *,
     coordinator: AgentCoordinator,
diff --git a/tests/test_execution.py b/tests/test_execution.py
index 4d3b5d25..94dccc1f 100644
--- a/tests/test_execution.py
+++ b/tests/test_execution.py
@@ -168,6 +168,55 @@ async def test_spawn_child_agent_respects_child_limit(
     assert set(coordinator.parent_of) == {"root", "child-1"}
 
 
+@pytest.mark.asyncio
+async def test_concurrent_spawn_child_agent_reserves_limit_atomically(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Any
+) -> None:
+    coordinator = AgentCoordinator()
+    await coordinator.register("root", "strix", parent_id=None)
+
+    async def _noop_start(**_kwargs: Any) -> None:
+        return None
+
+    monkeypatch.setattr("strix.core.execution._start_child_runner", _noop_start)
+
+    def _factory(**_kwargs: Any) -> object:
+        return object()
+
+    async def _spawn(name: str) -> dict[str, Any]:
+        return await spawn_child_agent(
+            coordinator=coordinator,
+            factory=_factory,
+            agents_db_path=tmp_path / "agents.db",
+            sessions_to_close=[],
+            run_config=cast("Any", object()),
+            max_turns=1,
+            interactive=False,
+            parent_ctx={"agent_id": "root"},
+            name=name,
+            task="do more recon",
+            skills=[],
+            parent_history=[],
+            max_child_agents=1,
+        )
+
+    async with coordinator._lock:
+        tasks = [
+            asyncio.create_task(_spawn("extra-a")),
+            asyncio.create_task(_spawn("extra-b")),
+        ]
+        await asyncio.sleep(0)
+
+    results = await asyncio.gather(*tasks)
+
+    assert [result["success"] for result in results].count(True) == 1
+    assert _child_count(coordinator) == 1
+
+
+def _child_count(coordinator: AgentCoordinator) -> int:
+    return sum(parent_id is not None for parent_id in coordinator.parent_of.values())
+
+
 @pytest.mark.asyncio
 async def test_concurrent_reserve_claims_yield_single_root() -> None:
     coordinator = AgentCoordinator()