OpenRouterLabs/spawn — reverse-engineered prompt

Reverse engineered prompt

Build me a CLI called Spawn that lets me launch different AI agents on different cloud providers with one simple command. If I just run spawn, it should open an interactive picker. If I run something like spawn claude sprite or spawn codex hetzner, it should launch directly. I want it to support passing a prompt or prompt file, dry run mode, headless mode with JSON output, choosing region, machine size, model, and loading options from a JSON config file. All model access should go through OpenRouter.

It should also feel practical day to day, so add commands to show the full agent and cloud matrix, list agents and clouds, track history, show parent and child spawn trees, rerun the last spawn, check live status, delete servers, repair an existing VM, and link an already running VM by IP. Include optional setup steps like GitHub, browser, and messaging integrations where relevant, plus a fast mode and a few beta optimizations. For GCP, make secure boot on by default with a way to turn it off. Look up current cloud and OpenRouter docs if you need to.

Want more depth? Deep Reverse