Why use <<<<<<< (7-8 characters) instead of simple XML like <search> for the search and edit functionality?

[akhilsonga]

https://github.com/Kilo-Org/kilocode/blob/main/src/core/diff/strategies/multi-file-search-replace.ts

Hello Devs, I am trying to learn how KILO code works, as a user it is amazing and I have a question:

I've been looking at the diff strategy code and I'm curious about the design choice for the marker format.
Is this because of Model Training data familiarity(Is it because LLMs have seen this Git merge conflict syntax more frequently in their training data compared to simple XML tags?)

Doesn't this format potentially confuse smaller models when editing files? They might mistake it for actual merge conflicts or struggle with the exact character count.

Or is it for token optimization? I noticed that 7-8 < characters are tokenized as a single token (in openai tokenizer) is it becasue of that?

Why not XML like other tags?

old code here new code here

Would love to understand the reasoning behind this design decision!

• Akhil