Primitif mines the skills behind your team's wins and shares them with every assistant.
Your know-how builds up as you work, so new engineers ramp up on it instead of starting from scratch.
curl -fsSL https://primitif.ai/install | sh
Just keep coding the way you already do, in Claude Code, Codex, or your terminal. Quietly, right on your own machine, we look back over your work and turn the things you figured out into reusable skills. You don't have to tag anything or remember to save it, it just happens as you go. Nothing new to learn, and nothing ever leaves your laptop.
Every skill and rule lives in one tidy home in your own repo, version-controlled right alongside your code. From there it's ready for each assistant in its own format, like Claude's .claude/skills and Codex's AGENTS.md, so there's no copy-pasting between tools. You write it down once, and it just works everywhere, today and with whatever assistants you add later.
As soon as a skill proves useful, it shows up in every teammate's assistant, right where they're already working. One person's hard-won fix quietly becomes the whole team's default. So nobody has to solve the same problem twice, and new folks pick things up without digging through old chat threads.
We keep an eye on how each skill does in real work, then polish the ones that help and quietly drop the ones nobody reaches for. The dashboard shows you what's getting used and what it's saving you. Your team's shared knowledge doesn't just pile up, it keeps getting sharper the more you all work.
Free for individuals.
One per-engineer plan adds team sharing, skill evals, and the dashboard.
More than 100 engineers? Let's chat →
Just a little info about your session, never your code.
Each skill is a plain markdown file you can open and read any time, and it all lives in your own repo. Your code and your ideas stay right where they belong.
Nope, and we really watch this. Most sessions add nothing at all. We only keep a skill when the work was tricky and worth reusing, and we keep each one short. They're just markdown in your repo, so you can delete any that have outstayed their welcome, and the dashboard quietly points out the ones nobody uses.
Your set stays small and handy, so each session stays light.
A small helper model does, automatically. After each session it reads over the work and only keeps something when it's tricky and likely to come up again. Everyday stuff and one-offs are skipped, so most sessions save nothing. Anything it keeps is plain markdown in your repo, so you can always read, edit, or delete it.
Want a person to okay a skill before the team gets it? That's coming soon.
Those make skills only when you run them, for one tool, then stop. Primitif quietly works in the background, keeps only the lessons worth keeping, and carries each one across every assistant you use. People have tried the idea before, but nobody made it automatic and able to work with every tool.
That's the part we got right.
You can, for one file, one tool, one moment. Primitif keeps it up to date for you and carries it across every assistant, so you're not rewriting the same lesson three times by hand. And on a team, one person's lesson quietly reaches everyone else.
Very little, honestly. It runs on a small, cheap model right on your own machine, and only the finished skill ever leaves. It's a tiny slice of what your real coding costs, and those tokens show up in your dashboard like any other, so there are no surprises.
Claude Code and Codex today. You capture a skill once, and we shape it into each tool's own format for you, so there's no copy-pasting between assistants. Cursor is coming next.
Yes, always. They're plain markdown in your own repo, yours to keep with or without us. Leave whenever you like and everything stays right where it is, still usable by any assistant.
No lock-in, ever.
Connect a repo in minutes.
Your whole team gets sharper from the first session.