alex-reysa/singular-lite ? reverse-engineered prompt

Reverse engineered prompt

Build me a bash and Python tool for orchestrating AI coding agents across a software repo.

I want it to manage three levels of work, one top level controller, area planners, and worker agents that run in isolated git worktrees on their own branches. It should keep track of leases for active tasks, recover stale work, merge finished work back safely, and write a simple human readable snapshot of the repo state after each cycle.

Please make detached dispatch the default, so the main command returns quickly while workers keep running in the background. I also want a gate and audit flow after each worker finishes, so test results can be checked and failures can be classified as retry, amend scope, escalate, or park. Include the repo setup flow, doctor checks, migrate support, and sensible config files and templates so a user can run it in a fresh repo.

If you need current details, look up the latest docs online and make it work cleanly from the command line.