get gpt to generate a dark package by using github's "code tree api" (via dark code-gen'd wrapper - find fsproj, get text of .fs files in order, concat, -> dark) on existing F# packages
- getting code form github
- https://docs.github.com/en/rest/git/trees?apiVersion=2022-11-28#get-a-tree
- https://docs.github.com/en/rest/repos/contents?apiVersion=2022-11-28#get-repository-content
- (there are a bunch of projects in this space - list them out, review them, see what would make sense in Dark)
- basic process
- parse xml -> dark (libxml)
- github package to...
- view/traverse files
- find fsproj
- extract list of files to concat
- concat files
- turn into prompt
-
dark
- iterate w/ parser
- fork canvas/experiment/package, with AI
- feed conversation with details of some intentions w/ fork diffs
- extend this into the idea of templates?
- try with some F# projects first (more similar langauge, and .fsproj is "is order")
- then typescript
- deal with nonlinear dependencies:
- read these files. what dependencies do these have? -> store the dependencies somehow
- sort the files based on the dependency graph, loosely
- then try to concat and convert
- I suspect that manipulating the order of files will matter a bit, in producing good generated output