Schedule Definitions

schedules holds the agent's schedules: what fires when, and where it goes. Each is the Agent State Protocol (ASP)'s schedule without its identity, which the agent's saved record keeps. In an agent's saved configuration these entries are under behavior. A standalone spec can define a recurring job like this.

schema = 1
[[schedules]]
name = "Morning inbox"
when = { kind = "cron", cron = "30 8 * * 1-5", zone = "Europe/Lisbon" }
message = "Summarise new client mail."
to = { kind = "owner", connection = "telegram" }
enabled = true

Destinations and Cadences

to can name an owner's direct connection or an exact thread, using its complete connection/chat identity. Without a connection, an owner destination uses the CLI inbox. An every cadence has positive seconds and an anchor in Unix seconds. An at cadence has one Unix instant. Two schedules of one name and malformed destinations are refused.

Identity

Which schedule each is to its runner is the agent's saved record's (~/.lnk/safe/agents/<agent>/schedules.json), by its name: its id, its revision and the native job it came from (origin), kept when migration or a harness switch carries one in. One the record lacks is spec:<name> at revision 1. A changed definition is published at the next revision, so the scheduler tells a change from a retried import.

Checked Reference

The field names below come from link_agent_spec::Schedule. Optional values remain unset until the feature resolves its environment or inherited settings.

root: enabled, message, name, to, when
to: connection, kind, thread
when: anchor, at, cron, kind, seconds, zone