I've been using multiple AI coding assistants (Claude Code, Cursor, Codex, Copilot, Gemini CLI, Windsurf, etc.) over the past few months, and I noticed the same problem everywhere.
Every assistant has its own way of defining skills, rules, or prompts.
Claude Code → Skills
Cursor → Rules
GitHub Copilot → Instructions
Codex → AGENTS.md
Windsurf → Workflows
Others → Yet another format
If I build a useful skill for one assistant, I have to rewrite and maintain it separately for every other assistant.
That feels a lot like JavaScript before npm or containers before Docker.
So I started building Kitbash.
It's not another AI coding assistant.
The goal is to create an open standard for portable AI skills.
Some ideas I'm exploring:
📦 Write a skill once and compile it for multiple AI coding assistants.
🔒 Versioned installs with lockfiles instead of copying prompt files.
🧪 Testable skills with evaluation suites.
🔐 Permission manifests and context budgets.
🔄
Discussion
Get the discussion rolling
A single comment can start something great.