kasetto

An extremely fast AI skills manager, written in Rust.

Name comes from the Japanese word カセット (kasetto) - cassette. Think of skill sources as cassettes you plug in, swap out, and share across machines.

Highlights

• Declarative - one YAML config describes your entire skill setup. Version it, share it, bootstrap a whole team in seconds.
• Syncs skills from GitHub, GitLab, Bitbucket, Codeberg/Gitea repositories or local directories including GitHub Enterprise and self-hosted GitLab.
• 21 built-in agent presets: Claude Code, Cursor, Codex, Windsurf, Copilot, Gemini CLI, and many more.
• MCP server management: declare MCP servers in the same config and Kasetto merges them into each agent's native settings file.
• Tracks every install in a local manifest - knows what changed and why.
• --dry-run, --json, and --verbose flags for scripting and CI.
• Ships as a single binary - install as kasetto, run as kst.

Why Kasetto

There are good tools in this space already - Vercel Skills installs skills from a curated catalog, and Claude Plugins offer runtime integrations. Both work well for one-off installs, but neither gives you a declarative, version-controlled config.

Kasetto is a community-first project that solves a different problem: declarative, reproducible skill management across machines and agents.

• Team consistency - commit a YAML file, everyone gets the same skills.
• Multi-source - pull from GitHub, GitLab, Bitbucket, Codeberg/Gitea repositories and local folders in one config.
• Agent-agnostic - one config field switches between 21 agent environments.
• Traceable - every install is tracked, diffable, and inspectable.
• CI-friendly - --json output and non-zero exit codes for automation.

Inspired by uv - what uv did for Python packages, Kasetto aims to do for AI skills.

Install

Standalone Installer

macOS and Linux:

curl -fsSL https://raw.githubusercontent.com/pivoshenko/kasetto/main/scripts/install.sh | sh

Windows:

powershell -ExecutionPolicy Bypass -c "irm https://raw.githubusercontent.com/pivoshenko/kasetto/main/scripts/install.ps1 | iex"

By default the binary is placed in ~/.local/bin. You can override this with environment variables:

Homebrew

brew install pivoshenko/tap/kasetto

Cargo

cargo install kasetto

From Source

git clone https://github.com/pivoshenko/kasetto && cd kasetto
cargo install --path .

Getting Started

1. Sync from a remote config or a local file:

# pull a shared team config from a URL
kst sync --config https://example.com/team-skills.yaml

# or use a local file
kst sync --config kasetto.yaml

That's it. Kasetto pulls the skills and installs them into the right agent directory. The next time you run sync, only what changed gets updated.

2. See what's installed:

kst list      # interactive browser with vim-style navigation
kst doctor    # version, paths, last sync status

Commands

kst init

Generates a starter kasetto.yaml in the current directory.

kst init [--force]

kst sync

Reads the config, discovers skills, and makes the local destination match.

kst sync [--config <path-or-url>] [--dry-run] [--quiet] [--json] [--plain] [--verbose] [--project | --global]

Missing skills are reported as broken but won't stop the rest of the run. The exit code is non-zero only for source-level failures.

kst list

Shows skills and MCP servers from the lock file(s). Without --project or --global, both scopes are merged so you can tell global and project installs apart (scope is shown per row / in JSON).

kst list [--json] [--quiet] [--plain] [--project | --global]

In a terminal (and without --plain), this opens an interactive browser — Skills and MCPs tabs with detail panes. Navigate with j/k, switch tabs with Tab or h/l, scroll with PgUp/PgDn, jump with gg/G. Use --plain, set NO_TUI=1, or pipe stdout for a plain text listing.

kst doctor

Prints local diagnostics: version, lock file path, installation paths, last sync time, and any failed skills from the latest run.

kst doctor [--json] [--quiet] [--plain] [--project | --global]

kst clean

Removes all tracked skills and MCP configs for the given scope.

kst clean [--dry-run] [--json] [--quiet] [--plain] [--project | --global]

kst self update

Checks GitHub for the latest release and replaces the current binary in-place.

kst self update [--json]

kst self uninstall

Removes installed skills, MCP configs, Kasetto data, and the binary.

kst self uninstall [--yes]

kst completions

Generates shell completion scripts.

kst completions <shell>

Supported shells: bash, zsh, fish, powershell.

Configuration

Kasetto looks for kasetto.yaml in the current directory by default. Point it at a specific file or URL with --config, or run kst init to generate a starter.

# Choose an agent preset (single or multiple)...
agent: codex
# agent:
#   - claude-code
#   - cursor

# ...or set an explicit path (overrides agent)
# destination: ./my-skills

# Install scope: "global" (default) or "project"
# scope: project

skills:
  # Pull specific skills from a GitHub repo
  - source: https://github.com/org/skill-pack
    branch: main
    skills:
      - code-reviewer
      - name: design-system

  # Sync everything from a local folder
  - source: ~/Development/my-skills
    skills: "*"

  # Pin to a git tag or commit
  - source: https://github.com/acme/stable-skills
    ref: v1.2.0
    skills:
      - name: custom-skill
        path: tools/skills

# MCP servers (optional)
mcps:
  - source: https://github.com/org/mcp-pack
  - source: https://github.com/org/monorepo
    path: mcps/my-server/pack.json

Supported Agents

Set the agent field and Kasetto figures out where to put things.

Don't see your agent? Use the destination field to point at any path.

Private Repositories & Enterprise

Set an environment variable and private sources just work — no login command, no credentials file:

Kasetto auto-detects GitHub Enterprise for any hostname with an owner/repo path, and GitLab self-hosted when the hostname starts with gitlab..

skills:
  # GitHub Enterprise
  - source: https://ghe.example.com/acme/skill-pack
    skills: "*"

  # Self-hosted GitLab (nested groups supported)
  - source: https://gitlab.example.com/team/ai/skills
    skills:
      - code-reviewer

The same tokens apply when you fetch a remote config via --config https://....

Roadmap

• Agents management
• Hooks management
• Your idea? Open an issue

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

Licensed under either MIT or Apache-2.0, at your option.