Looking at the compiler, it doesn't make much sense. You have an AST that is rather minimal and I can't seem to find an IR. There are files, such as cron.rs, which have very little code relating to what I would expect, and the tests seem to test very little that is worthwhile.
Some of these are absolute monsters, like this one: https://github.com/usedowe/dowe-lang/blob/main/crates/compil...
The codebase seems tightly coupled and to cover specific cases, as if something was off and more features were requested. I think a 12k line file is evidence of that.
Perhaps it is a useful tool for you but I would have trouble using something like this with just my quick skim because it seems a mile wide and an inch deep.
Also, because of the sheer size of it, it would be extremely difficult to have an agent properly comprehend much if anything. Which I would want in order to audit the internals.
The language itself seems minimal, but then I look at the skills, like this one for example: https://github.com/usedowe/dowe-lang/blob/main/skill-data/do...
That is a lot of tokens. The problem here is that you have a compiler; why are you encoding standards as a prompt instead of as compiler errors?
You can use linting and static analysis to force conformance with agents instead of suggesting with a prompt.
I wish you luck with the project. I'll have to pass, too much code for me to follow when I could simply have an agent directly write the html now days.