CodeX is actually a foreign employee The GPT 5.3 used by CodeX has strong capabilities, but its Chinese is really poor. It is recommended to treat it as a foreign IT employee from India. So, I'm doing this now: one ️⃣ Communicate ideas with Gemini 3.1 Pro For example, if I want to achieve dynamic cooling duration and dynamic opening amount, I first ask if the idea of Gemini 3.1 Pro is reliable and feasible. Provide a detailed feasibility analysis report. two ️⃣ Let Gemini 3.1 Pro release AI PROMPT I told Gemini that I plan to have CodeX perform specific programming tasks and provide a detailed AI PROMPT. Gemini provides a highly executable PROMPT with a thorough understanding of the context and core objectives. All in English, native AI language. three ️⃣ CodeX encoding CodeX, who works tirelessly, received the task and completed its assigned tasks. four ️⃣ code audit Currently, I will perform two code audits: 1. Open a new window using the Code Audit Skill I installed for CodeX, responsible for auditing its newly written code (not in the original window) 2. Let Gemini 3.1 Pro audit the code again (it has contextual information, it knows the specific target) After both audits were approved, git commit + push， Remote update. At present, it appears that this process is being executed quite smoothly. You no longer have to endure CodeX's headache inducing curry flavored Chinese (Gemini 3.1's Chinese language is that's great), nor do you have to worry about leaving vulnerabilities due to poor communication or AI code being unaudited. Of course, if you have Opus 4.6, replacing Gemini 3.1 Pro may yield better results. But who wouldn't give us a chance with Claude, the title maniac