sleuth-io/sx

What is sx?

sx (pronounced ess‑ex) is a tool for sharing AI “skills” and related assets – custom prompts, rules, agents, commands, hooks, and even tiny Model‑Context‑Protocol (MCP) servers – across a team. It treats a regular synced folder (Dropbox, Google Drive, OneDrive, etc.) or a Git repository as a vault that stores those assets together with a manifest (sx.toml) and a lock file. The vault becomes the single source of truth, and the sx app or sx CLI can:

Add assets to the vault (sx add …).
Install them for a specific scope – an organization, a repo, a path, a team, a bot, or an individual user (sx install …).
Track who is using what and how many tokens are spent (sx stats).
Audit every change (sx audit).
Expose the vault to web‑based AI assistants (Claude, ChatGPT) via a lightweight WebSocket relay (sx cloud …).

All of this works without any dedicated server – the shared folder itself is the infrastructure, and the CLI can be scripted or run in CI pipelines.

Who might use it?

Role	Why it helps
Non‑technical teammate	Uses the graphical app to drag‑and‑drop a skill and share it with the rest of the team – no Git or terminal needed.
Engineer / DevOps	Automates publishing and installing assets from scripts or CI (`sx add`, `sx install`, `sx vault copy`).
Team lead / manager	Gets adoption dashboards (`sx stats`) and an immutable audit trail (`sx audit`) to see which skills are actually used.
AI‑tool integrator	Generates the correct configuration files for Claude Code, GitHub Copilot, Cursor, Gemini, Kiro, etc., so the same skill works everywhere.
Product owner of an autonomous‑agent platform	Packages an entire agent (prompt + skills + rules + hooks) once and deploys it unchanged to any supported client.

How does it work?

Vault – a folder (local, synced, or Git) that contains:
- sx.toml – a manifest listing every asset, its version, and the scopes it should be installed to.
- sx.lock – a per‑user resolved view of the manifest (similar to package‑lock.json).
- .sx/ – audit logs, usage logs, and generated plugin descriptors.
CLI / App – both read/write the same configuration, so actions performed in the GUI appear instantly in the terminal and vice‑versa.
Scoping – when you run sx install <asset> --org or --team <name> the tool writes the appropriate files into the target AI client’s configuration directory, translating the generic asset definition into the client‑specific format.
Cloud relay – sx cloud serve runs a small process that forwards MCP requests from web‑based assistants (Claude, ChatGPT) to the local vault over a WebSocket, keeping the vault data private.
Go library – the same logic is exposed as pkg/sxvault, allowing other Go programs (e.g., the hetchy sandbox runner) to read/write vaults directly.

What can you actually put in a vault?

Asset type	Typical content
Skill	Prompt template, system messages, optional example inputs.
Rule	Lint‑style guidelines, file‑type restrictions, auto‑formatting policies.
Agent	A named bot definition that bundles a prompt plus the skills/rules it needs.
Command	Slash‑command strings that can be invoked from a client UI.
Hook	Scripts that run on lifecycle events (e.g., before a Claude session starts).
MCP Server	Small HTTP/WebSocket service exposing a model‑context protocol for custom data sources.
Plugin	A packaged Claude Code or Codex plugin manifest generated automatically from the vault.

How to get started (quick‑start)

# Install the CLI (macOS / Linux)
brew tap sleuth-io/tap && brew install sx
# or via the one‑liner script
curl -fsSL https://raw.githubusercontent.com/sleuth-io/sx/main/install.sh | bash

# Initialise a vault in a synced folder
sx init --type path --path ~/Dropbox/sx-vault

# Add a skill (drag‑and‑drop in the app or use the CLI)
sx add ~/.claude/skills/my‑skill

# Install it for the whole team (org‑wide scope)
sx install my‑skill --org

# See who is using it and token consumption
sx stats --since 7d --json

The same steps can be performed from the desktop app, which bundles the CLI automatically.

Limitations / Open questions

Area	Current state
RBAC / change‑request flow	Planned (roadmap entry).
Enterprise‑grade hosting	Large teams can use the hosted Skills.new service; self‑hosting is limited to the folder‑based models.
Supported AI clients	Broad (Claude, Codex, Cursor, GitHub Copilot, Gemini, Kiro, etc.) but some clients only support a subset of asset types (e.g., Gemini JetBrains – rules only).
Cross‑platform consistency	The app and CLI share a single config, but the bundled CLI does not self‑update; a separate install is needed for automatic updates.

TL;DR

sx is a zero‑infrastructure, version‑controlled vault system for AI‑related assets. It lets any team – technical or not – store, version, scope, and audit skills, agents, rules, commands, hooks, and tiny MCP servers, and automatically installs them into every supported AI assistant (Claude, Copilot, Cursor, Gemini, etc.). The tool is available as a cross‑platform desktop app and a Go‑based CLI, with a Go library for deeper integration.