Some thoughts on the architecture, FYI

1. Security/Logic: Why bundle OpenClaw inside the Sandbox? It feels counter-intuitive. OpenClaw should be the brain, and the sandbox should be the workspace.

2. Burn Rate: At the current rate, maintaining an active sandbox for a month looks expensive. AI projections show it hitting $100+, which is way above the $5 base fee.

3. Optimization: If we treat the Worker as the brain and decouple it from the environment, can we trigger heartbeat tasks via cron? We could spin up environments on-demand and use idle_timeout to kill them when not in use to save costs.

4. Also for scalable consideration, that's the core value of cloudflare based openclaw

The issue names no files, tests, or entry points. Start by turning the Worker and sandbox separation, heartbeat cron behavior, on-demand environments, idle_timeout, and cost concerns into an agreed scope; done is a maintainer-approved design or a concrete follow-up implementation issue.

Written by the indexing model from the issue text.