refactor(cron): keep occurrence authority fixed after admission (#201377) Problem: a running cron occurrence was aborted when its owner's role, device or membership changed (base: "Session event execution aborted"), because the restored operator authority subscribed to those changes and the occurrence re-checked its marker on every effect. Fix: a cron occurrence checks its owner's authority once when it is restored and starts; the running occurrence keeps that authority until it ends. Each new occurrence, queued child and replayed follow-up re-checks once at its own start (`reauthorize`, the person's current authority, never socket liveness). Job removal, explicit Stop, cancellation and release still end the occurrence. Refusals keep their specific reason instead of a generic "abandoned"/sign-in message. User impact: editing a job or the owner's role while an occurrence runs applies to the next occurrence instead of killing the current one; a revoked owner's next occurrence is refused with a clear reason. Proof: real Gateway from source with a real model: owner demoted while the occurrence's exec sleeps aborts it on base and finishes it on head; the next occurrence is refused on both. Regression tests cover first-start reauthorization, closed-connection starts, Stop, revocation, refusal reasons and the rewritten in-flight expectations. Builds on #201334. Co-authored-by: Ayaan Zaidi <hi@obviy.us>
refactor(cron): keep occurrence authority fixed after admission (#201…
Find Your Interest
- When your AI agent publishes internal data
- OpenAI connects the Dots
- 4 Steps to Changing People's Perceptions
- The Roundup by ByteByteGo
- Intelligence analysis platform for AI Agents (~OpenClaw)
- AI Engineer Interview Questions: What I Ask After the Right Answer
- WWDC 2026 roundup
- Last Week (Month) in AI #347 - Nvidia's spy, Anthropic's $1.4T IPO, AI's blue wave
- The AI is coming
- Exploring the TextAttack Framework: Components, Features, and Practical Applications





