Cool use cases of programming around AI

[moodymudskipper]

to start :

• go through the git log and build a table on information about commit, for a given branch and for each commit, list affected files, affected functions, (affected type of files too) types of changes among :
  • add new code
  • remove code
  • refactor code (keep functionality without adding a feature)
  • add feature
  • input validation
  • rename entities
  • reordering args
  • add comment
  • update comment
  • documentation
  • Formatting

We should be able to run as a recommit, to make sure we don't mix too many things, or to run globally, maybe updating a global database.

[moodymudskipper]

We could have a ask_commit() function in the package in fact, it seems that openai is fast enough to be practical when using context_diff() (not a perfect name btw, maybe we need context_uncommitted() or something).

ask_commit() proposes a git command in the terminal (in the style of git add . && git commit -m "Your commit message"). unless all = TRUE it decides if we commit everything at a time or if we commit in an order that makes sense to the AI.

The system message can be changed through options but by default it provides instruction not to commit unrelated changes together, or mix formatting changes with functional changes for instance.

Maybe it can work through a review process as in ask_in_place()where we can accept or reject (and we can still copy tweak and paste if we want)

[moodymudskipper]

Other use case: fact check documentation to see if it's still up to date.

We loop through pages of documentation (+ README and vignettes) and see if there's anything not up to date anymore, if there is we report with suggested in_place changes. Using ask_in_place() on the full repo might be enough, but we still need a good prompt.