stop-hook-must-check-for-actual-stop-event-not-message-content
The Stop hook condition should only evaluate session completion when the agent has actually initiated a stop action, not on every assistant message turn. The current hook prompt does not distinguish between ‘agent sent a short waiting message’ and ‘agent is stopping’ — both trigger the same evaluation path. Fix: the hook should exit immediately with APPROVE if no stop() call or agent-initiated termination is detectable in the transcript tail.