Should we add a `CLAUDE.md`?

[grantmcdermott]

I use AI pair programming every day. But, truthfully, I've hardly used it for tinyplot development. That's partly b/c the main code logic predates the arrival of (good) AI assistants, and partly b/c these assistants weren't always reliable for base R plotting and complex project context.

Still, it might be useful to provide the scaffolding for a consistent AI experience, especially w.r.t. repo structure and guiding outside contributions.

@vincentarelbundock I'm curious about your take here, since I see that you have CLAUDE.md files in other projects.

(I will say that I use SKILLS.md for my work / everyday projects and these have been a gamechanger for me.)

[vincentarelbundock]

I haven't really used SKILLS.md yet, but I plan to invest in understanding it soon.

Yeah, sure, we could check in a CLAUDE.md. But TBH, I'm not sure it's super necessary. The code base is small enough that /init will generate a new one at very low cost and almost instantly.

Perhaps it can still be useful to have it to provide guidance, style, etc.

[grantmcdermott]

Cool, cool. Do you have any advice on "auto-populating" a template that we can manually tweak, or non-obvious things that you'd include based on your experience elsewhere?

I haven't really used SKILLS.md yet, but I plan to invest in understanding it soon

I basically think of them as a way to instantiate different personas that aren't project-specific or dependent. For example, switch from an "r-expert" persona (that understands your personal programming and syntax prefs, favourite libraries, common use and problem patterns, etc.) to a "reviewer" persona (that you might have trained on your previous referee reports to identify common themes of things you look for, etc.). I recently gave an internal presentation demonstrating how I use them. Might see if I can cleave the sensitive parts and make public.

[vincentarelbundock]

Sounds very interesting. i'd for sure take a look a those slides.

Cool, cool. Do you have any advice on "auto-populating" a template that we can manually tweak, or non-obvious things that you'd include based on your experience elsewhere?

Not really. TBH, I haven't tweaked mine much beyond /init and saying things like: "never change unit tests unless I tell you to explicitly."