What is franke_skills?

franke_skills is a tool that lets a capable AI assistant, the one you are chatting with, hand off the actual coding work to separate AI coding tools running in the background, while it stays focused on planning and reviewing. The idea is that the assistant you talk to is good at breaking down a task and judging finished work, while the actual line by line coding is often better done by dedicated coding tools running on their own, with their own memory of the task, their own logs, and their own way of handling failures. The main piece included here is a single command line tool that wraps three different AI coding assistants, Codex, Claude Code, and Grok, behind one shared set of commands. Instead of learning a different way to start, check on, and collect results from each of these tools, you use the same handful of commands no matter which one is doing the work underneath. Compared to just running one of these coding tools directly in a loop, this project adds a few things. If a worker gets stuck or needs more information partway through, that shows up as a clear status you can check on, instead of the process silently failing or hanging. Checking in on running work brings the tasks that need attention to the top of the list, rather than making you scroll through everything. And a code review step is built in from the start, checking finished work both against general coding standards and against the original task description, with two separate checks run side by side and then combined into one judgment. Getting started means installing the skill for whichever coding tools you already use, which copies files into your project so your assistant can find and use them. It requires Python 3.10 or newer and at least one supported coding tool already installed. Once set up, you start a background coding task with one command, go do something else while it runs, check status later, then review and collect results, always reading the actual code changes yourself before trusting them. The project also lets you choose how much reasoning effort each backend uses for a given task, translating a single setting into the right option for whichever tool is doing the work. It is written in Python and released under the MIT license.