pejmanjohn/slot-machine ? reverse-engineered prompt

Build me this open source skill called Slot Machine.

I want a tool I can run from Claude Code or Codex that takes one feature request and launches several independent implementations in parallel, each using a slightly different approach. Then each version should be reviewed by a separate agent that looks for real bugs, and a final judge should decide whether to pick one winner, combine the best parts into a better result, or reject everything if none are good enough.

Please set it up so the main orchestration lives in SKILL.md, with the longer examples and reference material kept under references/, and make it work cleanly for both Claude and Codex style invocation. Include a simple way to configure the slots in a project file like AGENTS.md or CLAUDE.md, and make sure the workflow leaves clear result artifacts with what each slot did, what each reviewer found, and why the final choice was made.

If you need to check the current docs for Claude or Codex behavior, look them up online first.