🪴 Brain

Vercel Skills

4 min read

Vercel Skills

vercel-labs/skills — CLI for the open Agent Skills ecosystem. The command npx skills add <repo> installs skills from any repo (GitHub, GitLab, local path) into any of 50+ supported coding agents. It is Vercel’s answer to fragmentation — every agent has a different path for skills (~/.claude/skills/, ~/.cursor/skills/, ~/.config/opencode/skills/…), and Skills CLI abstracts this through symlinks or copies.

An implementation of the standard described at agentskills.io. Skills discovery via skills.sh.

🔗 Links

Description

Download or use

# Zero install — npx
npx skills add vercel-labs/agent-skills
 
# Full GitHub URL
npx skills add https://github.com/vercel-labs/agent-skills
 
# A specific skill from a repo
npx skills add https://github.com/vercel-labs/agent-skills/tree/main/skills/web-design-guidelines
 
# Local path
npx skills add ./my-local-skills
 
# Other commands
npx skills list                  # list installed
npx skills find typescript       # interactive or keyword search
npx skills update                # update to latest
npx skills remove <skill>        # remove
npx skills init my-skill         # generate a SKILL.md template

🗒️ Description

🧩 What it solves

Every coding agent has its own convention for skills:

AgentProject pathGlobal path
Claude Code.claude/skills/~/.claude/skills/
Cursor.agents/skills/~/.cursor/skills/
OpenCode.agents/skills/~/.config/opencode/skills/
Codex.agents/skills/~/.codex/skills/
GitHub Copilot.agents/skills/~/.copilot/skills/
Gemini CLI.agents/skills/~/.gemini/skills/
…(50+ agents)…

The CLI auto-detects installed agents on the machine and installs to all of them at once. Symlink mode keeps a single canonical copy, the agent points to it — updating one skill propagates everywhere.

🧩 Project scope vs global

ScopeFlagLocationUse case
Project(default)./<agent>/skills/Committed with the project, shared with the team
Global-g~/<agent>/skills/Available everywhere

🧩 SKILL.md format

---
name: my-skill
description: What this skill does and when to use it
metadata:
  internal: true   # optional, hidden unless INSTALL_INTERNAL_SKILLS=1
---
 
# My Skill
 
Instructions for the agent...

Required: name (lowercase, hyphens), description. Skills are generally compatible across agents, but some features are agent-specific (allowed-tools, context: fork, hooks — full matrix in the README).

🧩 Discovery and plugin marketplace compat

The CLI searches for skills in 50+ standard locations (skills/, .claude/skills/, .cursor/skills/, etc.) and also parses Claude Code plugin manifests (.claude-plugin/marketplace.json, .claude-plugin/plugin.json). That makes Vercel Skills compatible with the Claude Code plugin marketplace — you can consume skills from other people’s plugins through npx skills.

🧩 Bulk and CI/CD

# Everything to everything, no prompts
npx skills add vercel-labs/agent-skills --all
 
# Specific skills to specific agents
npx skills add vercel-labs/agent-skills --skill frontend-design --skill skill-creator -a claude-code -a opencode
 
# CI-friendly (no prompts)
npx skills add vercel-labs/agent-skills --skill frontend-design -g -a claude-code -y

Telemetry is off in CI automatically. DO_NOT_TRACK=1 optional.

✍️ Reasoning for

For me this is a complement to what I already have in Brain — my workflow skills (ingest, compile, reindex…) live in .claude/skills/ and are specialized for this vault. Vercel Skills won’t replace that, but it gives me a cheap way to consume community skills (frontend-design, skill-creator, web-design-guidelines from vercel-labs/agent-skills) without copy-paste.

The value is mostly in three things:

  1. Multi-agent install — when I test the same skill on Claude Code and Cursor (I have both) with a single npx skills add, I don’t copy by hand
  2. Plugin marketplace compat — compatibility with Anthropic plugins without vendor lock-in
  3. npx skills init — generates a SKILL.md template, saves boilerplate

Weak point: npx overhead on every install (downloads the CLI). For mass installs, global is better — npm i -g skills probably also works, although the README recommends npx.

Alternatives considered

  • Agent Skills (Anthropic native) — works only for Claude Code, no multi-agent
  • Karpathy Skills / single CLAUDE.md — always-on instructions, not load-on-demand
  • Awesome Claude Code — curation without an installer, manual copy-paste
  • Manual git submodules — works, but no multi-agent path translation
  • Claude Code plugin marketplace — Anthropic-only ecosystem, vendor lock-in

🔗 Resources

  • Agent Skills spec: https://agentskills.io/
  • Skills directory: https://skills.sh/
  • Agent docs links (per agent): in the README’s “Related Links” section — 30+ entries
  • Env vars: INSTALL_INTERNAL_SKILLS, DISABLE_TELEMETRY, DO_NOT_TRACK
  • Hermes Agent — TUI/messaging agent compatible with the same agentskills.io standard
  • Agent Zero / Space Agent — agentic frameworks using the open SKILL.md
  • Superpowers — methodology stack distributed as a plugin marketplace

Template: tool

Graph View

Backlinks