caponetto/ai-dev-orchestrator ? reverse-engineered prompt

Build me a local first orchestration tool for AI coding agents in TypeScript.

I want to define a workflow, then have the app run it from start to finish, sending tasks to different agent roles like planning, implementation, review, verification, and judgment. It should handle gates for approval and clarification, keep track of artifacts and run history, and recover cleanly if something crashes or gets interrupted. I also want a simple CLI for starting runs, checking status, listing past runs, resuming failures, approving gates, and inspecting artifacts.

Please include a web dashboard too, so I can see active runs, browse artifacts, review findings, and check token usage by role. Make it feel robust and reliable, with local execution, checkpoints, and clear state tracking. If you need to look up current docs online for the agent CLI integrations, go ahead.