Top AI Skills Top AI Skills
Guides

AI Agent Skills FAQ: Installation, Compatibility, and Troubleshooting Guide

· 6 min read · ★★★★★

Full disclosure: I’ve been using skills daily since early 2026 across Claude Code, Cursor, and Codex. I’ve installed, tested, and broken enough skills to have opinions. This FAQ is based on real experience — not theory.

Frequently Asked Questions

What are AI agent skills and how do they work?
AI agent skills are reusable knowledge packages that give your coding agent domain-specific expertise. When you install a skill, your agent loads structured reference files — API docs, best practices, code patterns, and troubleshooting guides — into its context. Instead of guessing or hallucinating, your agent answers from bundled reference material. Skills are installed via the skills.sh CLI with npx skills add <owner>/<repo>, and your agent picks them up on the next interaction.
How do I install skills for Claude Code, Cursor, or Codex?
Installation is the same across all agents. Run npx skills add <owner>/<repo> in your terminal. For example: npx skills add confident-ai/deepeval installs the DeepEval evaluation skill. npx skills add antfu/skills installs the entire set of frontend skills (Vite, Vue, Vitest). No restart needed — your agent picks up the new skills on the next prompt. You can verify installation by running npx skills list to see what’s currently loaded.
Which AI agent versions support skills?
Skills work with any agent that supports the skills.sh protocol. As of mid-2026, the confirmed-compatible agents are: Claude Code (latest), Cursor (0.45+), Codex (all versions), GitHub Copilot (1.200+), Windsurf (all versions), Cline (all versions), Gemini Code Assist, AMP, Antigravity, ClawdBot, Goose, Kilo, Kiro CLI, Nous Research, OpenCode, Roo, Trae, VS Code (with appropriate extensions), and Zed. If your agent can run npx commands, it likely works.
How do I uninstall or update a skill?
To uninstall, run npx skills remove <owner>/<repo>. To update a skill to the latest version, run npx skills update <owner>/<repo>. You can also update all installed skills at once with npx skills update --all. Skills are versioned based on the GitHub repo’s main branch — updates pull the latest reference files. Check npx skills list to see current versions.
What's the difference between a skill and a plugin?
Skills are knowledge packages — they add documentation, best practices, and reference code to your agent’s context. They don’t execute code or modify your editor. Plugins (or extensions) are executable code that add new features to your editor or CLI. Skills work alongside plugins: the plugin provides the interface, the skill provides the domain knowledge. On skills.sh, all packages are skills that follow the Agent Skills Specification.
Can I create my own skill for my team?
Yes. The Agent Skills Specification is open. Create a GitHub repo with a skills/ directory containing markdown or JSON reference files. Add a skill.json manifest in the root describing the skill’s metadata, compatibility, and entry points. Then install it with npx skills add your-org/your-repo. Teams use this to bundle internal API docs, coding standards, and deployment guides into skills all team members can share.
Why isn't my skill showing up after installation?
This usually happens for one of three reasons: (1) the repo doesn’t follow the Agent Skills Specification (missing skill.json or incorrect directory structure), (2) your agent needs a new session to pick up the skill (try restarting the agent or starting a new conversation), or (3) the skill is installed but your agent isn’t triggering on the right keywords (try being explicit: ‘using the [skill name] skill, how do I…’). Run npx skills list to confirm the skill is installed, then check the repo has a valid skill.json file.
Do skills work offline?
It depends on the skill. Most skills store reference files locally after installation, so the knowledge is available offline. However, skills that reference live docs (like Mastra’s remote docs fallback) may need an internet connection. Skills installed from GitHub repos are cached locally and don’t require internet to read — only to install or update. For fully offline setups, check each skill’s documentation for local-first support.
How do I find the right skill for my use case?
Browse the skills.sh directory at skills.sh — it has 9,500+ skills categorized by topic (React, Next.js, Design, Mobile, Agent Workflows, Databases, Testing, Marketing, and more). You can filter by agent type and sort by installs. For curated recommendations, check topaiskills.com for skill intros and tutorials with real-world testing results.
Can I install multiple skills from the same repo?
Yes. When you install a repo with npx skills add <owner>/<repo>, all skills in that repo are loaded. For example, npx skills add antfu/skills installs Vite, Vue, Vitest, and any other skills in that repo. If you only want specific skills from a repo, check if the repo supports granular installation — some repos let you install individual skills with npx skills add <owner>/<repo>/<skill-id>.

Getting Started with AI Agent Skills#

If you’re new to skills, here’s the short version: skills make your agent smarter about specific topics. Instead of your agent guessing how to use a framework or tool, it reads from structured reference files that come bundled with the skill. The result? Less hallucination, more working code.

The ecosystem on skills.sh has grown to over 9,500 skills with more than 865,000 total installs as of July 2026. That’s not a side project anymore — it’s how developers are augmenting their AI coding agents.

Quick Setup#

# Install a skill
npx skills add mastra-ai/skills

# List installed skills
npx skills list

# Update all skills
npx skills update --all

# Remove a skill
npx skills remove mastra-ai/skills

That’s it. No config files, no environment variables, no restart needed. Your agent knows about the skill on the next prompt.

Which Skills Should I Install First?#

If you’re just getting started, here’s a practical order:

  1. Your framework’s skill — Vite for frontend devs, Mastra for AI app builders, GSAP for animation work
  2. An SEO skill — catches issues before they hit production, works in the background
  3. A testing or evaluation skill — DeepEval for LLM apps, Playwright for E2E tests
  4. A Google Workspace skill — automates repetitive tasks like email parsing and spreadsheet updates
  5. A scraping skill — Firecrawl for research and data extraction

Install all of them — they don’t conflict. Your agent picks the right skill for the task automatically.

Common Pitfalls#

After months of daily use, here’s what I’ve learned the hard way:

Skills need explicit prompting sometimes. Your agent won’t always reference a skill unless you ask it to. Try phrasing prompts like “Using the [skill name], how do I…” or “Check the [skill name] reference for…”

Not all skills are created equal. Some repos have minimal documentation. Check the GitHub repo before installing — if the skills/ directory is sparse, the skill won’t add much value.

Skills can conflict with each other. This is rare, but if you get inconsistent answers, try running npx skills remove <repo> on your least-used skill and see if quality improves.

The skills ecosystem is moving fast. Skills you installed three months ago might have better versions now. Run npx skills update --all weekly.

Want to read in-depth reviews of specific skills? Check out our skill directory for detailed intros. For a curated list of the most useful skills, see our AI agent skills roundup. And if you’re just learning how to set up your first skill, our installation guide walks through it step by step.