[aysha-afrah26]

When a Sleep node is halted, onHalted() cancels its timer through TimerQueue::cancel(), which does not wait: it re-queues the handler so the worker thread runs it shortly after with aborted=true. That handler cleared timer_waiting_ unconditionally, even though onHalted() had already done it. If the Sleep is started again before the worker gets to the canceled handler, which is what happens with tree.haltTree() followed by a new tick or with a parent that halts and re-ticks it, onStart() sets timer_waiting_ for the new timer and the stale handler then resets it, so the next onRunning() returns SUCCESS. In practice a finished after about 20 ms in 44 of 50 halt/restart cycles on macOS, which matters for trees that rely on Sleep to wait before acting. The aborted path now returns without touching the node, and only a timer that really expired clears the flag and emits the wake-up signal. The change stays inside sleep_node.cpp with no header, layout, API or ABI change, and Decorator.SleepRestartedAfterHaltWaitsFullDuration fails on master at the first iteration and passes with the fix; the full ctest suite is green locally apart from cmake_subproject_version, which fails the same way on the untouched tree because of my local SDK, and clang-format is clean (clang-tidy-21 was not available here).