stop-hook-fires-on-every-assistant-turn-not-just-stop-calls
The Claude Code Stop hook event fires after every assistant message turn ends (when the agent is in a waiting state for user input), not only when the agent explicitly invokes stop(). This caused dozens of false-positive evaluations mid-brainstorm where the hook kept asserting ‘no stop action has been initiated.’ The hook’s condition ‘before allowing the agent to stop’ does not mean ‘only on explicit stop()’ — it triggers on every natural turn boundary.