Token/Code compression?

[TomLucidor]

I remembered that Caveman skill exists, and since a lot of open-weight models are only 131K, I kinda wonder if there are ways to make things faster/portable/small https://github.com/JuliusBrussee/caveman/discussions/97

[TomLucidor]

For code there is the following (Ponytail acts like senior dev who refuse to write too much code) DietrichGebert/ponytail#17 (comment)

[kashim93]

can You use ponytail on top of OmO ? should I run /ponytail everytime ?

[TomLucidor]

That is the question I am asking, like shouldn't some agent just have these thoughts by default?

[TomLucidor]

Bonus observation (and maybe skills can be automative as well) https://github.com/JayPokale/RDXmin