> loaded into the working model — the original file is untouched and will stay that way
Commentary text like this, once described here on HN as essentially the chat context leaking into the application, is absolute anathema to me. Something about it makes me irrationally annoyed and immediately not want to use the software.
I think, to me, it signals the pinnacle of vibe-coding in the sense that I can't imagine that the output was looked at closely and the copy still remained. When my own agents try to write stuff like that, I remove it immediately.
Like I said, maybe irrational.
> Corruption needs something to touch, and we touch nothing.
It always reads to me like some high school essay where the student has nothing more to say but a wordcount requirement they still need to hit.
I wonder if this is caused by so much of the training data is the web, where millions of articles were written to hit some word count. It's just following what it was trained on.
Nah, not irrational. No application should show the logic at the forefront. It's just claude-ese convention from RLVR
It’s very annoying. Claude specifically will tend to treat minor clarifying details in a prompt as major features.
I’ve also noticed it likes to spam a UI with explanatory text for every control. Almost like it wants to do tooltips, but doesn’t understand the exact UI for them.
You're definitely not wrong. It's the new code smell.