[NaySurGithub]

What

system.beforeEvents.watchdogTerminate was bound but never fired. It now fires when a pack's script runs past the script watchdog deadline:

• the interrupt handler finds the hung pack from the running frames (the innermost frame whose file sits under a pack root)
• once the hung code has unwound, that pack alone gets a WatchdogTerminateBeforeEvent with terminateReason: "Hang" and cancel: false, run in restricted execution like every before event
• one event per arming: the code that trips right after an expired deadline is not a new hang

How it was checked

Client 26.60 in IDA, the scripting watchdog's handler:

• a hang report fires the event with reason index 0 (Hang), a stack overflow report with index 1 (StackOverflow); the event carries the plugin it is about
• after the event, when cancelling is allowed and the event was cancelled, nothing else happens; otherwise it logs an unhandled critical exception and, when shutdown is enabled, shuts the server down
• both switches come from the world's script settings. Their values on a dedicated server are not in the client binary, so this PR does not shut anything down and cancel has nothing to stop; the interrupted call is aborted exactly as before

Not covered: StackOverflow. This server's engine reports a stack overflow as an ordinary catchable RangeError with no hook to tell it from a script's own, so there is nothing reliable to fire on.

Test: script-api-watchdog: a pack hangs in a timeout; it receives exactly one Hang event, keeps running afterwards, and a second pack subscribed to the same signal receives nothing.

Not built locally (syntax-checked with clang++ -fsyntax-only); CI runs on the branch.

AI disclosure

Code, research in the game binary and this description: Claude Opus 5.5 (claude-opus-5-5).