> Want to keep LLM .md files in a separate overlay, only make them visible on request? Also easy. CRDT gives the freedom in splitting and joining along all the axes.

I now have a bunch of layers of text / markdown: system prompts, AGENTS.md, SKILL.md, plus user tweaks or full out replacements to these on every repo or subproject.

Then we want to do things like update the "root" system prompt and have that applied everywhere.

There are analogies in git, CMS templating systems, software package interfaces and versioning. Doing it all with plain text doesn't feel right to me.

Any other approaches to this problem? Or is Beagle and ASTs and CDRTs really onto something here?

https://www.unison-lang.org/docs/the-big-idea

If it were intended to be general replacement for general purpose version control systems, I am not sure how storing AST is better than storing the original plain text files since the transformation from text to AST might be lossy. I might want to store files with no AST (e.g. plain text files), files with multiple AST (e.g. polyglots), multiple files with the same AST (e.g. files to test different code layout), broken AST (e.g. data files to be used as test cases). These use cases would be trivially supported by storing the original file as is, whereas storing any processed form of the file would require extra work.

are these asts fully normalized or do (x) and ((x)) produce different trees, yet still express the same thing?

why change what is being stored and tracked when the language aware metadata for each change can be generated after the fact (or alongside the changes)? (adding transform layers between what appears and what gets stored/tracked seems like it could get confusing?)

https://news.ycombinator.com/item?id=47294924

Hopefully this is the path to protentional editing/editors - the underlying code is an AST, but what you see, and edit, is the human-friendly text representation of that AST. Of course, you need a solid transform not just from text/PL repr to AST, but also AST to PL, possibly keeping local metadata relevant to the second part (e.g. where to put non-generated-by-default-from-ast whitespace and other formatting), so this might call for new (or modified) prog-langs for this explicit purpose?

From "Large Language Models for Mathematicians (2023)" (2025) https://news.ycombinator.com/item?id=42899805 :

> It makes sense for LLMs to work with testable code for symbolic mathematics; CAS Computer Algebra System code instead of LaTeX which only roughly corresponds.

> Are LLMs training on the AST parses of the symbolic expressions, or token coocurrence? What about training on the relations between code and tests?

There are already token occurrence relations between test functions and the functions under test that they call. What additional information would it be useful to parse and extract and graph rewrite onto source code before training, looking up embeddings, and agent reasoning?