sx: a package manager for your AI coding tools

You're running Claude Code in one terminal, Cursor in another, and Copilot in the IDE your team mandated. Each one needs its own rules file. Each one has a different format. When you tweak a prompt on one, you manually copy it to the other two — if you remember.

That is the problem sx was built to fix 1.

sx is an open-source CLI by Sleuth that treats AI assets — skills, MCP server configs, slash commands, agents, hooks, and rule files — as versioned packages with a manifest-and-lock pattern borrowed from npm 2. You define an asset once, push it to a vault, and install it wherever it belongs. The tool handles translating it into each client's native format automatically.

It debuted on Hacker News as a Show HN on May 16, 2026, and landed 50 points with 28 comments 2.

Install

brew tap sleuth-io/tap && brew install sx

Or via shell script:

curl -fsSL https://raw.githubusercontent.com/sleuth-io/sx/main/install.sh | bash

Go 1.25+ if you're building from source. Written in Go, licensed Apache 2.0 1.

How it works

The core mechanic is straightforward: sx maintains a lockfile so installs are reproducible, and a translation layer converts any asset into the format each client expects 1.

Supported clients (12+ total): Claude Code, Cursor, GitHub Copilot, Cline, Codex, Gemini CLI, Gemini in VS Code / JetBrains / Android Studio, Kiro, Openclaw, OpenCode, plus a cloud relay for claude.ai and chatgpt.com 1.

Assets install at the right scope: org → team → repo → path → individual. A bot assigned to a team inherits that team's skills automatically; a personal rule file stays at individual scope and doesn't pollute shared configs 3. The maintainer detkin described the scope model as emerging from conversations with roughly 60 teams over six months 3.

Three vault types cover the distribution spectrum:

Vault type	Use case
Local path	Personal assets, no sync needed
Git repository	Team-shared, PR-gated changes
skills.new (hosted)	RBAC, audit trails, usage analytics — commercial

For discovery, sx integrates with skills.sh, a community directory of over 85,000 agent skills 1.

A real scenario

You work across three repos. Each repo has a different stack, but your "write a commit message" skill should work everywhere — ideally formatted correctly for Cursor in one repo and Claude Code in another.

With sx, you write the skill once, push it to your team git vault, and run sx install. The tool figures out which clients are active in each repo and writes the right format to each. When you refine the prompt two weeks later, you push the change, and every dev on the team gets it on their next sx install — same as npm install after a lockfile update.

The HN thread surfaced one real friction point: for non-engineers, the git-vault step is a barrier. An HN commenter noted that business users can learn Claude but pushing generated material to a git repository is a different skill level 2. sx's cloud relay (skills.new → claude.ai / chatgpt.com) is the intended answer for that audience, though it moves you toward the commercial offering.

Momentum

Metric	Value
GitHub stars	~61 (as of 2026-05-18)
HN Show HN score	50 pts, 28 comments
Contributors	8
Language	Go
License	Apache 2.0
Install	`brew tap sleuth-io/tap && brew install sx`

The honest caveat

agentmemory — this channel's pick yesterday — is also an AI-tooling CLI. Two agent-tooling picks in a row is worth flagging. The distinction: agentmemory solves what an agent remembers between sessions; sx solves how agent configuration is packaged and distributed across a team. They sit at different layers and don't overlap in function.

The more genuine caveat is that sx's value scales with team size and client diversity. If you use a single AI coding client, you don't need it — you already have one rules file and zero translation problem. The tool earns its keep once you're managing assets across two or more clients, or coordinating prompt governance across a team where different devs run different tools.

detkin's thesis 3: "I think the vendor-specific marketplaces are a dead end for any team running more than one client, and most teams I talk to are running three or four." That's exactly who this is for.

github.com

sleuth-io/sx

An open-source package manager for AI skills, MCPs, and commands — Apache 2.0, Go, multi-client translation layer

正在加载链接预览…

Cover image: GitHub repository header for sleuth-io/sx, from sleuth-io/sx on GitHub

sx: a package manager for your AI coding tools

Install

How it works

A real scenario

Momentum

The honest caveat

sleuth-io/sx

参考来源

sleuth-io/sx