@huoshan007: Guys, many projects aren't poorly written by Codex; it's that you haven't thought it through yourself. When I encounter vague requirements, I prefer to start with this: 'Don't write code yet. Ask me five key questions consecutively, point out contradictions directly; finally, compile a requirements list and wait for my confirmation before proceeding.' This is very effective. It forces 'I kind of want' into 'specific...'
Summary
The tweet suggests that when using Codex, first have it ask key questions to clarify vague requirements, compile a requirements list, and then start coding to improve efficiency and reduce rework.
View Cached Full Text
Cached at: 08/30/26, 02:19 PM
Hey folks, often it’s not that Codex writes poor code — it’s that you yourself haven’t clearly thought things through.
When I run into vague requirements, I prefer to drop this prompt first:
“Don’t write code yet. Ask me five of the most critical questions, point out any contradictions directly, then compile everything into a requirements checklist. Wait for my confirmation before you start.”
This works surprisingly well.
It hammers down vague ideas into “exactly what needs to be done.”
For big projects, use it first as a product manager, then as a programmer. Less rework, and way more effective than just rushing it to move faster.
Similar Articles
@alin_zone: Claude Code and Codex, stop choosing between them. Instead of agonizing over which to use, let them divide the work—one writes, one reviews. Working together is smoother than using either alone. This post covers my own workflow: Claude Code handles writing, Codex handles reviewing. The entire collaboration takes about five steps, and you make the final call.
Introduces a workflow: using Claude Code for writing code and Codex for code review. The two complement each other more efficiently than using either alone. The entire process takes about five steps, with the final decision left to the developer.
@LukeDing14: Many people on X still don't know how to use Codex to complete projects. They find it strange that their Codex cannot write high-quality code or organize projects more efficiently. A month ago, the popular learn-harness-enginnering resource already covered many tips. Source link here ht…
LukeDing14 points out that many people still don't use Codex correctly for projects, and recommends the learn-harness-engineering resource along with his own harness file, sharing tips for efficient coding and project organization.
@SUOHA_AI: You must change how you use Codex — the results will be an order of magnitude better. 90% of people treat Codex like a god and go straight to /goal: help me write an app. In reality, it can hardly understand your specific needs. Share a trick used by OpenAI researcher @reach_vb: first let Co…
Share a tip to improve Codex usage: don't set the final goal immediately, but first let Codex do requirements gathering and research, then set a concrete executable /goal based on the findings, which can significantly improve code generation quality.
@xiangxiang103: Friends who just installed Codex, don't rush to use it—if you don't do these two things, it will only get dumber over time. This thread perfectly explains the "taming guide" I personally use: ① Let it first get to know you and your computer; list the 10 most important things to tell it in one sentence. ② Set up a dedicated workspace + two key documents (pitfall log / AGE…
This tweet provides a detailed Codex "taming guide", suggesting that you first let Codex get to know the user and the computer, set up a dedicated workspace and create a pitfall log and AGES behavior guidelines document to improve its usage and avoid getting dumber over time.
@ErwinWu000: Just one prompt, using Codex's native features to help even beginners do graph engineering — a foreman manages N underlings for me, and I just sit back and sip tea.
Author ErwinWu000 shares how to implement lightweight Graph Engineering using only Codex's native features, a single prompt, and stage documentation. Through a "general contractor" Agent managing multiple independent Worker tasks, it implements a staged multi-Agent workflow, and provides a ready-to-run template.