a5c-ai/babysitter ? reverse-engineered prompt

Build me a JavaScript tool called Babysitter that helps me run agent workflows in a safe, controlled way.

I want a command line app and plugin setup that can plug into different coding assistants, so I can define a workflow once and use it across supported tools. It should let me run steps in order, pause for human approval at checkpoints, enforce that each step passes before moving on, and keep a clear log of what happened so I can review it later. If something fails, it should stop cleanly and make it easy to resume.

Please include a good onboarding flow, install instructions, and examples for common use cases, plus a simple docs site or documentation pages. Also add tests so the core workflow behavior is reliable. If you need current docs for the supported harness integrations, look them up online first.