Short version: start with Anthropic's official library and one of the curated indexes, then add a token saver and a memory skill. Star counts below come from the GitHub API on 18 September 2026. Popularity tells you what people install; it is not a security review, so read a skill's files before you add it.

Official libraries
**anthropics/skills** is Anthropic's public repository of Agent Skills and the reference for how a good SKILL.md is structured. Read two or three before writing your own.
**addyosmani/agent-skills** is a set of production-grade engineering skills for coding agents, and the one people search for by name.
**google-labs-code/stitch-skills** pairs with Google's Stitch MCP server for UI generation.
Curated indexes
**ComposioHQ/awesome-claude-skills** and **VoltAgent/awesome-agent-skills** are the two big catalogues, the second collecting over 1,000 skills. Browse these before building something that already exists.
Token savers
**caveman** is built on one idea: fewer words, fewer tokens. **codeburn** is not a skill but a local tracker that tells you whether any of this is working.
Memory and planning
**claude-mem** captures what the agent learned so the next session does not start cold. **planning-with-files** keeps long tasks on track across context resets.
Design and behaviour
**ui-ux-pro-max-skill** and **taste-skill** both target the generic-looking interfaces agents tend to produce. **andrej-karpathy-skills** is a single CLAUDE.md distilled from Andrej Karpathy's coding guidance.
How to install a skill

- Pick the scope.
- Just you, every project: ~/.claude/skills/<name>/.
- The whole team, this repo: .claude/skills/<name>/.
- Copy the skill folder in.
- It must contain SKILL.md with frontmatter first.
- Check the description.
- Claude decides when to load a skill from its description.
- Invoke it or let it load.
- Run /skill-name, or let Claude pick it up when relevant.
Read before you run
Skills can ship scripts that execute on your machine. Open `SKILL.md` and every script before installing, and be wary of any skill asking for broad permissions through `allowed-tools`. The full, regularly updated list is in the Agent Skills directory.