Stop using abbreviations

Resolved 💬 5 comments Opened Aug 31, 2025 by lukaszciastko Closed Sep 3, 2025

💡 Likely answer: A maintainer (guidedways, contributor) responded on this thread — see the highlighted reply below.

What feature would you like to see?

The GPT-5 model is consistently using abbreviations in code, which makes the code unreadable. Please instruct it in the default prompt not to do this - it's a very bad (and annoying) coding practice.

Are you interested in implementing this feature?

_No response_

Additional information

_No response_

View original on GitHub ↗

5 Comments

guidedways contributor · 12 months ago

Same - it uses variables such as p or v etc. The solution seems to be to add explicit instructions in AGENTS.md and your desired coding standards (with examples, otherwise it still won't follow instructions). Annoying to hav to do this though as Claude Code never does this.

guidedways contributor · 12 months ago

It seems AGENTS.md is frequently ignored - seems to be a widespread CLI issue across Claude / Codex / Gemini.

Any way, it's back to ignoring code-style related instructions and is spitting out unreadably dense, short and abbreviated code with 1900s C style variable naming, one-line loops and if / else and the lot.

dedrisian-oai contributor · 11 months ago

Closing as this is a model behavior. As mentioned above, try adding to AGENTS.md (with several strong words :)

guidedways contributor · 11 months ago

@dedrisian-oai I've tried everything you can imagine in AGENTS.md it just doesn't help. How can we have the model deal with this better? Example of horrendous looking swift code:

public var shapeHint: ShapeHint = .rectangle { didSet { syncTextSystemToView() } }

public enum ShapeHint { case rectangle, roundedRect, triangleUp, triangleDown, diamond, ellipse, trapezoid(topInsetRatio: CGFloat) }

Who formats variables in one line? Codex does this to loops, if / else, functions, variables, just about everything.

guidedways contributor · 11 months ago

@dedrisian-oai separately: https://github.com/openai/codex/issues/2927

Also, after around 40% context usage it seems to begin ignoring whatever it consumed in AGENTS.md