skills mit Homebrew, Nix installieren

Projektlinks

• Homepage
• Paketmanager-Seite

Befehle und Aliase

• add-skill
• skills

Projektgeschichte

Vercel Labs created Skills as an open CLI and ecosystem for discovering, installing, using, updating, and removing reusable AI-agent skills. It accepts GitHub shorthand, GitHub and GitLab repositories, general Git URLs, local paths, direct `SKILL.md` files, archives, well-known providers, and skills.sh packs.

The ecosystem grew beyond installation into catalog search, anonymous install telemetry, popularity rankings, public and private packs, temporary one-shot skill use, and APIs for programmatic discovery.

Adoptionsgeschichte

The official README reports support for OpenCode, Claude Code, Codex, Cursor, and more than 70 additional agents. Its cross-agent installation design uses a canonical skill copy with optional symlinks or independent copies in agent-specific directories.

skills.sh adds discovery, install counts, rankings based on anonymous CLI telemetry, and shareable packs, turning the CLI into both a package installer and the client for a broader agent-capability catalog.

Wie es verwendet wird

Users commonly run `skills add` to inspect and install selected skills, `skills use` for temporary prompt generation or an interactive agent session, `skills find` for discovery, and `skills list`, `update`, or `remove` for lifecycle management.

Installation defaults to project scope; `--global` targets the user scope. Users may select particular agents and skills, install everything non-interactively, and choose symlinks or copied files.

The CLI stores global installation/update metadata in `.skill-lock.json`. If `XDG_STATE_HOME` is set, the documented implementation uses `$XDG_STATE_HOME/skills/.skill-lock.json`; otherwise it uses `~/.agents/.skill-lock.json`.

Warum Paket-Nerds sich dafür interessieren

For package enthusiasts, Skills applies familiar ecosystem mechanics—source resolution, project and global scopes, lock metadata, updates, removal, search, archives, Git URLs, and registries—to portable AI-agent instructions. Its interesting twist is that the installed artifact is usually a directory centered on `SKILL.md`, rather than a conventional executable library.

The package also exposes the interoperability problem directly: one source skill can be linked or copied into the layouts expected by many different agents, making it a compatibility layer as much as a downloader.

Zeitleiste

• 2025: skills.sh recorded early public agent-skill listings and began surfacing install activity.
• 2026: Official documentation described a mature catalog, packs, API, anonymous rankings, and support for a large range of agents.
• 2026: Lock-file schema version 3 tracked source metadata and full skill-folder hashes for update detection.

Related projects

• skills.sh: the official catalog, documentation, leaderboard, pack service, and API used with the CLI.
• Agent Skills repositories: source repositories containing one or more directories built around a `SKILL.md` manifest and instructions.
• Supported coding agents, including Codex, Claude Code, Cursor, and OpenCode, consume installed skills through their respective filesystem layouts.

Quellen

• Official CLI reference: https://www.skills.sh/docs/cli
• Official Skills documentation: https://www.skills.sh/docs
• Official lock-file implementation: https://github.com/vercel-labs/skills/blob/main/src/skill-lock.ts
• Official repository README: https://github.com/vercel-labs/skills#readme