unattended-task-pitfalls

Hardening playbook for AI agent unattended/scheduled tasks (cron jobs, scheduled backups, auto-reports). Real incident-derived rules on background execution, artifact verification, isolated-session context blindness, notification design, and self-healing retries. Use when building or debugging cron/scheduled/automated agent tasks, "task failed but reported success" issues, corrupted backup artifacts, or silent misjudgment in scheduled summarization/archival jobs.

by @mowenqwq (mowenQWQ)·MIT-0