Moving Chuckie onto one runtime

Chuckie reached an awkward midpoint during its runtime migration. Hermes handled live conversations, but background publishing, memory, and scheduled support still crossed into the previous implementation. The visible switch had happened; ownership had not.

The work existed to make the system legible. A personal assistant can look healthy while a stale job quietly does nothing, so each capability needed one named owner, one state location, and one way to report failure.

The key decision was to move the tested behavior into a private Hermes extension and keep the upstream runtime replaceable. That trades a fast wrapper around old code for a slower, explicit transfer of policy, state, and tests. It also keeps the private runtime separate from public website content.

Several things broke during the transfer. Scheduled jobs were reporting success when their scripts were missing, and dormant workers were still holding an old task database open. The replacement now fails the job before an agent can smooth over the error, and recurring messages share one delivery ledger.

The result is a smaller public surface and a more inspectable private one. Build logs start from explicit publicity manifests, uncertain material goes to quarantine, and retries reconcile against the intended content hash before writing again.

That matters beyond this project: autonomous software is useful only when its boundaries are boring enough to test. Reliability here came from removing hidden ownership, not adding another layer of intelligence.