wtf is subagentmaxxing?

subagentmaxxing lets a Claude Code agent hand off tasks to other AI coding tools, specifically OpenAI's Codex and the Cursor CLI, and treat them the same way it treats its own native subagents. You give it a prompt, pick which model should handle it such as Grok, Composer, or GPT, choose a working folder, and set how much freedom the subagent has to actually make changes, and it hands back one clean answer. You can also send the same prompt to several models at once and compare their answers side by side. The whole tool is a single Python command line program called subagent, with no extra dependencies to install, plus a matching Claude Code skill that teaches an orchestrating agent when to reach for it. The idea behind the project is that Claude is very good at planning and deciding what to do, but running the actual coding work through cheaper models like Cursor's Composer or xAI's Grok can be dramatically less expensive while still getting most of the quality, so pairing a strong planner with cheap workers can save a large amount of money on agentic coding tasks. Setup is a git clone followed by running an install script, which puts the subagent command on your machine and installs the matching skill. A doctor command checks that each backend tool is installed and properly logged in. You can run a single subagent, fan a prompt out across many models at once for a consensus answer, run tasks in the background and check on them later, or even keep the subagents running on a separate, more powerful machine and drive them remotely over SSH from your own laptop.