The idea: use your smartest model for intelligence and cheaper models for execution.

You can run it across your whole codebase or on your current working branch.

Every plan comes with audit, discovery, scope, execution, testing and stop conditions.

GitHub: https://github.com/shadcn/improve

@kapish_dima right. any codebase. not ui related. not frontend. a general coding skills.

@shadcn will fable run "/improve security"?

@shadcn Did I understand correctly that this isn't specifically a shadcn/ui improvement, but something for the codebase in general?

@ephraimduncan Yes.

@shadcn this is so good.

think I'll keep it in the toolbelt in general, I usually plan with an intelligent model and then send off to Composer 2.5 for cheap grunt work 🫡

@ludylops Yes. Run `/improve --issues` and it will create one issue per plan.

(By default plans are written under ./plans.)

@shadcn Amazing, will be using it today. Instead of saving the plan can I ask it to just open well structured issues on my repo?

@ephraimduncan Here's sample output.

@shadcn i am on it

@shadcn good thank you, i’m going to try it

@shadcn https://gist.github.com/OmerFarukOruc/753f95b1ac278b683be83ed26b3bcc1f

I have thought doing same and prepared with the Claude's Fable prompting guide. Maybe that /improve skill can be finetuned with those instructions.

https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5

@shadcn on it 👀

@shadcn This is exactly how I use Fable now, great job!

@shadcn where is your cape man? you saving out here 🔥👨🏾‍🍳

@shadcn / fixmyshit

@shadcn thankyou design engineers be like

@shadcn Went into X to look for a best practice with Fable. This was the first comment 🔥

The ROI in time you give me is insane 😅

@kostyniuk00 @shadcn bruv it’s a one liner…

@shadcn