Write an AI agent skill once and compile it for nine different coding assistants.
See a skill's per session token cost before installing it.
Review a third party skill's permissions and compiled output before trusting it.
| singhharsh1708/kitbash | cvlab-kaist/cats-plusplus-project-page | sebastiancarlos/hacker-fables | |
|---|---|---|---|
| Stars | 8 | 8 | 8 |
| Language | HTML | HTML | HTML |
| Last pushed | — | 2022-10-20 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | easy | easy | hard |
| Complexity | 2/5 | 1/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs Node 20+ for the npm install path, or Homebrew as an alternative.
Kitbash is a command line tool and open file format for writing AI agent skills once and running them across many different coding assistants, including Claude Code, Cursor, Copilot, Codex, Cline, and several others. Instead of maintaining a separate copy of the same instructions for every tool your team uses, you write a skill in Kitbash's own format and its compiler turns that single source into whatever format each target assistant actually reads. A skill in this format is a folder containing a manifest file describing its budget and permissions, an instructions file, and optional scripts and tests. What sets the compiler apart from a plain converter is that it measures how many tokens a skill will add to every single session before you even install it, since some assistants can load instructions lazily on demand while others must load everything upfront, which can multiply the ongoing cost many times over for the exact same skill. The tool surfaces that cost difference directly rather than hiding it. Getting started means running an install command, either from GitHub, a plain owner and repository shorthand, or a local file, followed by a compile command that generates the native output for whichever assistants you target. Before installing anything from someone else, a preview command can fetch and render a skill's compiled output, permissions, and token cost without actually installing it, and installation itself shows what a skill declares before writing any files. A lockfile records a content hash for each installed skill so later drift between what was reviewed and what is actually present can be detected. The project is released under the Apache 2.0 license and distributed through npm or Homebrew.
A compiler and package manager for AI agent skills, letting you write one skill and run it in Claude Code, Cursor, Copilot, and more.
Mainly HTML. The stack also includes Node.js, CLI, npm.
Use freely, including commercially, as long as you keep the license notice, Apache 2.0 also includes a patent grant.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.