martyaxel/ai-orchestrator-runbook

ai-orchestrator-runbook is a single markdown file, not really a piece of software, that sets up a particular way of working with two different AI coding assistants together. The idea is that Claude Code handles the thinking: it plans out the work, breaks it into pieces, hands tasks off, and checks the results, while a separate assistant called Codex, based on a model referred to as gpt-5.5, does the actual mechanical execution of those tasks. According to the README, this arrangement is meant to give strong quality results while relying mostly on your subscription plans rather than paying premium, pay per use prices for routine work. Using it is meant to be very simple. You open whichever AI coding agent you already use, such as Claude Code or Cursor, and either give it the included file, called runbook.md, or just tell your agent to fetch it directly from its raw GitHub URL and execute it. From there, the agent is expected to carry out the whole setup itself, only pausing to ask you to log into a service when needed, and the whole process is estimated to take about three minutes. To use this setup, you need an active Claude subscription, either Pro or Max, along with an active ChatGPT subscription, either Plus or Pro, since the actual intelligence and capability come from those subscriptions rather than from the runbook file itself. You also need an editor like Cursor or VS Code, and the README notes that the free tier of either is enough. The runbook itself includes a model routing table, saved as a CLAUDE.md file, that describes the cost, intelligence, and stylistic tradeoffs of different models, which lets the orchestrating assistant automatically choose an appropriate model for each task without you having to decide manually.