also having the two models work independently and asking Claude to analyze the differences and summarize them. Finally， manually adjusted to be used as prompts for GPT. How to use it? Download AGENTS.md and put it into your Codex home directory (defaults to ~/.codex。
I provided several small but representative development tasks for GPT and Claude to complete independently. Then， but it has recurring code style issues. This is likely a side effect of OpenAI's reinforcement learning stage. The goal of this project is to correct GPT's code style through prompts. How we did that? First。
Claude analyzed the differences in code style between the two and summarized a report. I then provided several new feature development tasks on top of the existing codebase， language-agnostic set of code guidelines， claude-your-gpt GPT is a smart model — it can solve problems and find bugs accurately， the two summaries are merged and distilled into a concise， unless you set CODEX_HOME) or project root. It also works with other AI coding tools that support custom instructions or system prompts. ，。