feat: add official Codex skill and harden task execution

Install and update artemis-use with task diagnostics and plugin guidance. Preserve cross-device file receipts, review canonical reports, guard stage transitions and reduce small history summaries. Improve exploration lifecycle, request recovery and browser live connections.

tests/unit/mcp/test_shared_backend.py

@@ -245,6 +245,45 @@ async def test_executor_cancellation_reaches_coroutine():
         await executor.stop()
 
 
+@pytest.mark.asyncio
+async def test_sync_tool_runs_once_off_both_event_loops():
+    import threading
+
+    executor = ToolExecutor()
+    executor.start()
+    caller = threading.get_ident()
+    calls = []
+
+    def operation(value):
+        calls.append(threading.get_ident())
+        return {"accepted": value}
+
+    try:
+        assert await executor.invoke(operation, "job") == {"accepted": "job"}
+        assert len(calls) == 1
+        assert calls[0] not in {caller, executor.thread.ident}
+    finally:
+        await executor.stop()
+
+
+@pytest.mark.asyncio
+async def test_sync_tool_error_does_not_repeat_the_side_effect():
+    executor = ToolExecutor()
+    executor.start()
+    calls = []
+
+    def operation():
+        calls.append("submitted")
+        raise ValueError("reply lost")
+
+    try:
+        with pytest.raises(ValueError, match="reply lost"):
+            await executor.invoke(operation)
+        assert calls == ["submitted"]
+    finally:
+        await executor.stop()
+
+
 def test_stale_pid_and_wrong_installation_are_not_reused(tmp_path, monkeypatch):
     monkeypatch.setattr(tempfile, "tempdir", str(tmp_path))
     directory = registry.registry_directory()