Scheduled daily digest

Problem

You want a daily check-in record — a compact summary of what tasks closed, what dispatches cost, and any alerts that fired in the last 24 hours. It should land on a shared thread at 09:00 local time every day without anyone triggering it by hand. The agent producing the summary can be an existing review agent; you don't need a new one.

Recipe

One scheduled trigger. Dispatches your review agent at 09:00 with the goal "produce daily digest".

# Create the scheduled trigger.
spl config add-trigger \
  --name "daily-digest-0900" \
  --schedule "0 9 * * *" \
  --target "did:sync:agent:director" \
  --goal "Summarise yesterday's closed tasks, dispatch costs, and alerts. Emit a KNOW on th_daily_digest with body.topic=daily_digest." \
  --budget 0.50 \
  --timeout 600 \
  --max-concurrent 1

# Verify.
spl config list-triggers

# See when it last fired and what it emitted.
spl config trigger-history --limit 10

The cron expression 0 9 * * * fires at 09:00 every day (minute=0, hour=9, any day of month / month / day of week). Standard 5-field cron. The trigger dispatches the target actor with the given goal, then emits a KNOW on th_engine_config capturing the trigger event. The agent's own work (fetching the data, composing the digest, emitting the digest record) happens on a dispatch sub-thread.

The agent should emit the digest to a stable thread — here we're using th_daily_digest. Create it once if it doesn't exist:

# Open the digest thread with a root INTEND so it has a name and a home.
spl intend "Daily digests" --thread th_daily_digest

Subsequent KNOWs from the agent land on that same thread and fold into a chronological log of daily summaries. spl thread records th_daily_digest reads the archive.

The trade-off

Cron triggers are instance-local. If you're running a fleet, the trigger fires on whichever instance owns the th_engine_config record (the coordinator in most deployments). If that instance is offline at 09:00, the digest skips — there's no makeup run. For a multi-instance fleet that must run the digest regardless, author the trigger on the fleet's coordinator thread and rely on the coordinator-redundancy story.

The goal text is the entire brief the agent sees. Rich briefs produce better digests. The goal above asks for a specific thread (th_daily_digest) and a specific body topic (daily_digest), which means you can CEL-match on the digest records later (e.g., route them through a notification webhook, alert on missing digests). A vague goal ("summarise yesterday") will work but produces less structured output.

Budget caps the dispatch. 0.50 USD on a competent reasoning agent is enough for a focused digest over ~50 closed tasks. Raise if the team is bigger; lower if you want to force conciseness. The trigger will emit TIMEOUT if the dispatch exceeds 600 seconds or the budget; the next day's 09:00 fire is unaffected.

See also

• Scheduled triggers guide — cron syntax, schedule + match combinations, history inspection.
• CEL expressions — if you want the trigger to also fire on record events instead of (or in addition to) cron.
• The Engine — CRON loop — what else runs on the CRON timescale.