Message Cleanup — daily housekeeping body

This is a scheduled-loop body skill, not a user-facing workflow. The message-cleanup canonical loop fires once a day (cron 17 3 * * *) and wakes a session to run this skill. The runtime is short: one CLI verb + a receipt log.

When this fires

The TUI cockpit registers this loop alongside cortex-mailbox-poll when the user toggles L on an instance for the first time, or by explicit empirica loop register --name message-cleanup. On fire, the AI sees a <task-notification> and loads this skill.

What to do

One command, then close out:

empirica message-cleanup --output json

The verb walks refs/notes/empirica/messages/ for any message whose expiry_at is in the past and removes them. JSON output shape:

{
  "ok": true,
  "dry_run": false,
  "removed_count": 12,
  "removed": [
    {"message_id": "...", "channel": "...", "subject": "..."},
    ...
  ]
}

If removed_count > 0, log a brief finding so the cleanup is visible in the project's audit trail:

empirica finding-log \
  --finding "message-cleanup: pruned <N> expired mesh messages" \
  --impact 0.2 --epistemic-source intuition

If removed_count == 0, no artifact needed — silent success is fine.

Then signal heartbeat and you're done:

empirica loop heartbeat message-cleanup --status ok --result \
  $([ "$removed_count" -gt 0 ] && echo found || echo empty)

What NOT to do

• Don't open an empirica transaction for this — it's a pure CLI

cleanup, no praxic decisions to gate.

• Don't run with --dry-run unless you're debugging. The loop's

scheduled fire IS the action.

• Don't escalate to user attention unless the verb errors. A scheduled

cleanup is supposed to be invisible.

Source

• Verb: empirica message-cleanup — handler at

empirica/cli/command_handlers/message_commands.py::handle_message_cleanup_command

• Underlying logic: empirica/core/canonical/empirica_git/message_store.py::cleanup_expired
• Catalog entry: empirica/core/cockpit/canonical_loops.py (this loop)