feat: add support for cursor cli #54
Conversation
|
@hugodutka @matifali PR ready for review. |
|
@hugodutka are we good to merge here? |
matifali
requested review from
hugodutka and
johnstcn
and removed request for
hugodutka
Co-authored-by: Cian Johnston <public@cianjohnston.ie>
Co-authored-by: Cian Johnston <public@cianjohnston.ie>
|
@johnstcn thanks for the review, made the changes. |
| if !strings.Contains(line, m) { | ||
| return currentIdx, false | ||
| } |
There was a problem hiding this comment.
I'm also relatively new to this codebase, so please let me know if I'm way off --
my understanding is that we want to match a "bottom line" of box-drawing characters.
This approach could potentially incorrectly match a line similar to:
Hello (╯▽╰) this is my custom agent
Would a better approach be to check that the string trimmed of leading and trailing whitespace starts or and ends with the given characters?
There was a problem hiding this comment.
check that the string trimmed of leading and trailing whitespace starts or and ends with the given characters?
makes sense, will add that.
Anyhow the logic above it
msg, msgLines, msgRuneLineLocations := normalizeAndGetRuneLineMapping(msgRaw)
userInput, _, userInputLineLocations := normalizeAndGetRuneLineMapping(userInputRaw)
userInputStartIdx := findUserInputStartIdx(msg, msgRuneLineLocations, userInput, userInputLineLocations)
if userInputStartIdx == -1 {
// The user input prefix was not found in the message prefix
// Return the original message
return msgRaw
}
userInputEndIdx := findUserInputEndIdx(userInputStartIdx, msg, userInput)
// Return the original message starting with the first line
// that doesn't contain the echoed user input.
lastUserInputLineIdx := msgRuneLineLocations[userInputEndIdx]
to get the lastUserInputLineIdx tries it's best to make sure we are getting the line that is right after the user input message ends. Can't think of a case where the next line to the user input (in case of gemini and codex in which the behaviour is known) is going to be anything other than the end line.
PS: This got me thinking, the matching logic was initially there because I didn't have access to the agent type in the RemoveUserInput, since I do that now, I think we can just skip the next line (since the behaviour is known ?), wdyt ?
ref: coder/registry#309