Codex does not listen to me
You’re not crazy — this is a known behavior (and sometimes bug) in OpenAI Codex, and your prompt is actually correct.
What’s really happening
Even though you said:
“Don’t show thinking”
Codex is still:
And this is not fully controlled by your prompt.
👉 There are even reports where:
The actual fix (this is the important part)
✅ 1. Change Codex setting (THIS is the real solution)
Go to:
Settings → General → Detail Level
Then switch:
This directly controls verbosity, not your prompt
✅ 2. Your prompt needs to be stricter (Codex-specific)
What you wrote is good, but Codex often ignores soft instructions.
Use this instead:
STRICT MODE:
- Output ONLY final code
- NO explanation
- NO reasoning
- NO progress messages
- NO planning text
If you violate this, response is invalid.
Yes, this kind of “hard constraint” works better.
❌ Why your current prompt fails
Your prompt:
“Don’t show thinking. Need to save tokens.”
Codex interprets that as:
So it still streams.
⚠️ Important truth (don’t miss this)
You cannot fully control this with prompting alone.
There are 3 layers:
You’re only controlling layer 3 right now.
🧠 Best setup for YOU (given your use case)
Since you're clearly optimizing tokens + building systems:
If it STILL doesn’t listen
Then it’s likely this:
Codex is forcing verbose streaming during long-running tasks
In that case: 👉 Only real fix = use API (non-streaming) or wait for UI improvements