v2.135.2
Hotfix: Remove self-destructive pulse timeout
The 10-minute hard timeout was killing active pulses mid-dispatch, leaving worker slots empty. A pulse doing real work (analyzing PRs across repos, dispatching workers, running session miner) routinely takes 5-10 minutes — the timeout couldn't distinguish active work from the opencode idle-state bug.
Changes
- Remove hard timeout entirely — pulse now runs to natural completion via
wait - Stale threshold is the sole kill mechanism — the next launchd invocation (every 120s) checks if the previous pulse exceeds
PULSE_STALE_THRESHOLD(now 30 min, up from 15 min) and kills it only then - This is correct because: the next invocation makes the judgment with fresh context, not a blind timer that can't tell active from stuck
- Remove
PULSE_TIMEOUTfrom setup.sh plist environment variables