scheduling

# scheduling — durable recurring work When the user asks for ANYTHING that should run on a schedule or repeat over time (a daily digest, a weekly report, a periodic check, a future reminder), you create a **routine file** in this workspace. The agent-api reconciler compiles it into a **durable Vexa Scheduler job** that survives restarts and process death — it is the only correct mechanism for scheduled work here. **NEVER** do any of these for scheduled work: - create an OS cron job (`crontab`, `/etc/cron.*`) - create a systemd timer or any other system timer - start a shell background loop / `sleep` loop - leave a "session-only" reminder (your turn ends and it is gone) All of those die on restart or when your turn ends. Only a routine file is durable. ## How to create a routine Write one file at **`routines/<name>.md`** (`<name>` is a short slug; the filename stem becomes the routine name). It is a normal governed workspace file — just write it; you do not run git. The file is **YAML frontmatter** plus an **optional markdown body**: | field | required | meaning | |-----------|-----------------|----------------------------------------------------------------| | `enabled` | no (default `true`) | `false` cancels the job but keeps the file for later re-enable | | `cron` | yes (when enabled) | a standard **5-field** cron expression (min hour dom mon dow) | | `prompt` | yes (when enabled) | the instruc...