Just FYI. The proofread breaks quotes. This has that way a while now.
Doesn’t really break the quote, only the preview visualization.
I’d say this, plus the the animated diff we have on the roadmap, means we should move to doing a diff without the cooking part of it.
What do y’all think @awesomerobot @keegan ?
Just to add to this…
We recently updated Discourse, and from what I recall, the plugin wouldn’t try to proofread quoted text (as it shouldn’t because technically it’s not really quoted anymore if it does).
Below is a screenshot…
@keegan just rolled out a new implementation, the diff is now applied to markdown and streamed ![]()
It works great for this case and even handles fixing typos in quotes ![]()
אני כאן כי אני לא חושב ש-proofread צריך לתקן חומר מצוטט (הוא לא מצוטט אם הוא “מתוקן”).
האם יש דרך לבקש מעבר לא לגעת בחומר מצוטט?
דרך אגב: עבודה נהדרת על השדרוג האחרון והפיכת פלאגינים רבים, כולל AI, כסטנדרט. אני מארח את עצמי והייתי צריך להסיר כמה פלאגינים כדי לשדרג שהיו כסטנדרט. (תודה על חוויה חלקה ששומרת על האימוג’ים שלנו). אני שמח שכל השאר שלא יודעים להוסיף פלאגינים מקבלים כעת את החוויה המלאה. בדיוק הגדרתי את חווית ה-AI אתמול וזה טוב.
I think we can do something here. It is somewhat tricky cause we need to hoist out content and then re-insert it into the post.
I certainly consider this a bug, will put a pr-welcome on this for now, I may get to it, but it is a bit tricky. The server side needs to partially parse the markdown, remove the quotes, give to proofreader and then add them back in, this may end up resulting in the proofreader getting an incoherent thing to proof read.
Tricky problem.
Is proofreading quotes actually a real issue
After all, the error should have been fixed by the quoted user in the first place. Anyway, as long as the AI doesn’t change the context itself — and it does/did make such changes sometimes.
I almost hate to say this out loud, but worrying about changes per se sounds almost like code-based behavior where the most horrible thing that can happen is if x=y changes to x=1 even when the latter one is true too ![]()
But sure. If a quote can stay untouched, just like a code block should be left alone, that would be nice or even great. But is it such a big thing that anybody should use a lot of paid hours to fix it?
Another sure. That is just my opinion and I know there are situations where a quote should be a 1:1 copy.
It can be a problem.. especially if ai makes a mistake with quoted religious texts like pāḷi material and makes a mistake doing it.
Or if the user forgot the special characters (which ai does a nice job of doing), then it really should not fix that, because the quote is the quote . He might lead to a small tangent by misspelling a pāḷī word, but if it is corrected, then it does not justify the rant for his miscommunication due to a spelling mistake.
So, I think it should preserve the quote.. it does not feel right for me to upgrade someone’s quote.
Excellent.. I updated earlier today before following up.
I look forward to keeping peoples bad “qotes.” ![]()
Testing below for fun:
We use Grok 4 Fast because it is cheap.. but considering we spend 3 Cents per month… We can look at other models if there are problems. We have a slightly custom prompt agent. I’ll test now since we are updated.
Any suggestions for ai engines?
We have a custom persona called pali_proofreader
You are a markdown proofreader and a Pāḷi scholar. Fix grammar, Pāḷi diacritic marks, and wrong spellings for a Theravāda discourse discussion website. Correct typos and phrasing issues but keep the user’s original voice. Do not touch quoted blocks (except to remove accidental extra line breaks from copy/paste). Preserve paragraph formatting. If nothing needs fixing, return the text unchanged. Remove unnecessary line breaks, especially from PDF copy/paste. Fix incorrect or missing diacritic marks. Output ONLY the corrected text, with no explanations, no notes, and no commentary.
Are you using examples? The default proofreader agent ships with 7 examples to help ground it, specially for quotes.





