I constantly hear developers who have tried AI saying "it's not there yet" when discussing leaning on AI for code. This is a workflow I've developed over the past 2-3 years and now it's formalized into a little project called: Wrinkl.
It's a tiny CLI + folder convention that:
scaffolds a .ai/ directory inside your repo (wrinkl init)
lets you spin up “feature ledgers” (wrinkl feature user-auth) where you jot down intent, edge cases, test plans, etc.
snapshots lean context files you can paste (or soon auto-feed) into your LLM so it stays grounded
adds a simple archive command when the feature ships, so your context window stays small
I constantly hear developers who have tried AI saying "it's not there yet" when discussing leaning on AI for code. This is a workflow I've developed over the past 2-3 years and now it's formalized into a little project called: Wrinkl.
It's a tiny CLI + folder convention that:
scaffolds a .ai/ directory inside your repo (wrinkl init)
lets you spin up “feature ledgers” (wrinkl feature user-auth) where you jot down intent, edge cases, test plans, etc.
snapshots lean context files you can paste (or soon auto-feed) into your LLM so it stays grounded
adds a simple archive command when the feature ships, so your context window stays small
Repo: https://github.com/orangebread/wrinkl (MIT)
Open to feedback and discussion!