fix: reliably dispatch and return background tasks

agent/task_worker.py

@@ -36,7 +36,7 @@ class TaskWorker:
 
         reg.add_step(self.task_id, "thinking", f"Starting: {self.description}")
 
-        client = AsyncOpenAI(base_url=base_url, api_key=api_key)
+        client = AsyncOpenAI(base_url=base_url, api_key=api_key, timeout=60.0)
         tools = self._load_tools()
 
         messages = [
@@ -45,6 +45,12 @@ class TaskWorker:
         ]
 
         max_iterations = 10
+        try:
+            await self._run_loop(client, model, tools, messages, max_iterations, reg)
+        finally:
+            await client.close()
+
+    async def _run_loop(self, client, model, tools, messages, max_iterations, reg):
         for iteration in range(max_iterations):
             if self._cancelled:
                 reg.fail(self.task_id, "Cancelled")
@@ -106,7 +112,7 @@ class TaskWorker:
         return (
             "You are Hope's research assistant working on a background task. "
             "Use your tools to gather information thoroughly. When you have "
-            "enough, provide a final answer in 2-4 sentences suitable for "
+            "enough, provide a final answer in 1-3 sentences suitable for "
             "speaking aloud. Plain text only, no markdown or formatting."
         )