Remote-Control Relay With MCP and Approval Gates

This build lab focuses on the local relay that lets an agent help with desktop tasks without becoming an uncontrolled operator. The goal is not to copy a private machine setup. The goal is to learn the architecture pattern well enough to build a small, classroom-safe version.

Computer-control agents need a relay that separates observation from action and requires approvals for risky clicks, typing, files, app launches, or shell commands.

Build the small version

1. 1Draw or write a control-flow diagram for screenshot, plan, approval, action, and audit logging.
2. 2Mark which parts are user-facing, which parts are internal, and which parts require approval.
3. 3Choose one low-risk workflow and implement only that workflow first.
4. 4Add one failure case before adding a second feature.
5. 5Write a short operator note: what the agent may do, what it must ask about, and what it must never do.

remote action loop:
  observe: screenshot or app state
  propose: model writes next action
  classify: low, medium, high risk
  approve: require human for medium/high
  execute: relay runs one allowed action
  audit: store action, reason, result
  stop: kill switch always wins

The big idea: approval gate is not decoration. It is part of the product architecture students need before an agent becomes safe enough to use with real people.