Short answer: Frontend Design + Impeccable for UI, Ponytail for leaner code, Caveman for fewer tokens, Superpowers for planning, Context7 for docs, Playwright MCP for testing.
A skill is a folder with a SKILL.md file: instructions, and sometimes scripts or reference files, that teach the agent one way of working. It is an open standard: Claude Code, Codex, Cursor, Gemini CLI and GitHub Copilot all read the same format. The agent reads only the short description at the start of a session and loads the full skill when a task matches, so installed skills cost almost nothing until they are used.
The easiest way to install one is npx skills add owner/repo, which puts it in the right folder for each agent you use. In Claude Code, skills also come bundled in plugins that you install with /plugin. An MCP server is different: it is a running program that gives the agent new tools, such as a browser or a documentation search, and it works in any agent that supports MCP.
The list above is short on purpose. Every pick fixes a specific problem I ran into while building with AI coding agents, and each one is open source, so you can read exactly what it tells the agent before you install it.