Avoiding doing the work

Resolved 💬 2 comments Opened Apr 18, 2025 by julianbcgc Closed Apr 19, 2025
💡 Likely answer: A maintainer (BadPirate, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

0.1.2504172351

Which model were you using?

o4-mini

What platform is your computer?

Kubuntu 24.04

What steps can reproduce the bug?

I am actually not really sure how to reproduce it yet. This is a defect that occurred where after some period of time codex would begin a strange mode where it perform a coding effort by planning it out, but no matter how many times I would tell it - CODE THE DAMN THING, for example - it just would keep coming up with reasons why its a big change and then give me examples of why, somehow refusing to actually make the change despite increasingly strong directions to get going. I wish I could offer more, but this is a weird behaviour I haven't seen in other CLI tools yet.

What is the expected behavior?

Well, it should just make the change.

What do you see instead?

_No response_

Additional information

_No response_

View original on GitHub ↗

2 Comments

BadPirate contributor · 1 year ago

I've also had trouble getting agent to properly "do the work" even with --full-auto and instructions to make changes, verify that they work and that lints pass, and DON'T COME BACK UNTIL ITS GOOD, or you are blocked -- Agent will follow instructions sometimes, ignore it other times and will provide me a list of commands for me to run and next steps for me to do (the same that I just gave to it).

fouad-openai contributor · 1 year ago

This seems like a model behavior issue, so I'd agree with @BadPirate and suggest prompting or using the new /compact command. o3 is less "lazy" and we definitely plan to make smaller models too! I'm going to close for now but will tag as model-behavior so we can track for the future.