Show HN: A scriptable text editor for LLMs (github.com)

3 points by dhamidi ↗ HN
Built this out of curiosity to see whether LLMs get better a editing if they express edits as a series of operations instead of a diff.

I don't have my answers yet, but actually building a PoC was the first step.

Claude seems to be happy to use the tool and write fake elisp just fine.

Think of it as sed without the `s`, and a tiny bit of Emacs instead.

What I'm looking for: fellow experimenters to get feedback and maybe develop the idea further.

The implementation is intentionally naïve.

Have fun!

1 comment

[ 3.0 ms ] story [ 14.1 ms ] thread
It's certainly worth exploring. Thee way most agents work right now bugs me. Sending huge diffs seems inefficient, uses a lot of tokens, and is prone to error. Another option that came to mind is working at the AST level, but that's a whole other can of worms to open.