Reverse engineered prompt

GitHub

Build me a Codex plugin in Python that lets me set up a simple multi model workflow for any Codex task. I want Codex to stay in charge, but be able to hand planning to a Planner, plan review to an Advisor, and implementation to an Executor. The Planner and Advisor should be optional, the Executor should be required, and the Advisor should be able to send the plan back for revisions until it returns PLAN_APPROVED, with a hard stop after a few review rounds.

Make it easy to configure with chat style commands like setup, status, disable, and a way to create reusable custom roles for models from other providers that are already authenticated. It should save the workflow and apply it to new tasks automatically, and also work with Codex Goals.

Please support using Claude Fable 5 as a special built in option for planning or review, with effort levels like low through max. Keep permissions and credentials untouched, and make sure saying no subagents prevents delegation. Look up current Codex plugin docs if needed.

Have a live product UI? Try website reverse