What a skill is, precisely
A skill is progressive disclosure for instructions. The model sees only the skill's name and description at first; when a task matches, it reads the full SKILL.md and any files the skill points to. That keeps context small while giving the agent deep, task-specific knowledge on demand — procedures, style rules, domain checklists, even executable helper scripts.
---
name: release-notes
description: Write release notes from the git log in this repo's house style. Use when asked for a changelog or release notes.
---
1. Run `git log --oneline <last-tag>..HEAD`.
2. Group commits by area (api, ui, infra). Drop chores.
3. One line per change, user-facing wording, no commit hashes.
4. Output in the format of CHANGELOG.md; do not edit the file unless asked.Where Claude Code finds skills
~/.claude/skills/<name>/SKILL.md— personal, every project..claude/skills/<name>/SKILL.md— committed with the repo, shared with the team.- Inside plugins — the distribution mechanism for bundles of skills, slash commands, subagents and MCP config.
The plugin marketplace
/plugin marketplace add anthropics/claude-code # register a marketplace (any git repo with a manifest)
/plugin install <name>@<marketplace> # install from it
/plugin # browse, enable, disable, updateA marketplace is just a repository with a plugin manifest, so teams host private ones the same way. Anthropic's official skills live at github.com/anthropics/skills — document skills (PDF, DOCX, XLSX, PPTX), a skill-creator skill, and examples that are worth reading before you write your own.
Skills people actually install
- Document skills — read and produce PDF, Word, Excel and PowerPoint files properly instead of guessing.
- Frontend design skills — a component library, tokens and layout rules so generated UI matches the product.
- Testing and review skills — the team's definition of done, encoded once.
- Deployment runbooks — the exact commands and checks, with the dangerous ones fenced off.
- Our curated list, with links and what each one is for: github.com/iskandaryv/awesome-claude-skills.
Writing a good one
- Make the description specific about *when* to use it — that is what the model matches on.
- Keep SKILL.md short and put bulk reference material in separate files the skill links to.
- Prefer scripts for deterministic steps; the model should call them, not re-derive them.
- Test it with
/skillsin a session and watch whether it triggers on the tasks you meant.
Skills on the gateway
Skills are files on your machine, read by the Claude Code runtime, so they are completely independent of which API endpoint is configured. Skill-heavy sessions are also long sessions — which is the case a flat-rate plan is built for.
Frequently asked questions
What is the difference between a skill and a plugin?
A skill is one capability (a SKILL.md folder). A plugin is a distributable bundle that can contain many skills plus slash commands, subagents and MCP servers, installed from a marketplace.
Are Claude Code skills free?
Yes. Anthropic's official skills and community collections are open repositories; skills cost tokens only when they are actually loaded into a session.
Where is the Claude Code skills marketplace?
There is no single store. Any git repository with a plugin manifest is a marketplace; register it with /plugin marketplace add. Anthropic's own repositories are the usual starting point.
Can I use skills with the Agent SDK?
Yes — the SDK runs the same runtime, so skills in the standard directories load the same way.
Run Claude Code on the gateway
Same models, two environment variables, credits at 7.69× face value — or a flat-rate unlimited plan.
Get an API key See unlimited plansAI Prime Tech is an independent API gateway and is not affiliated with, endorsed by, or sponsored by Anthropic. “Claude” and “Claude Code” are trademarks of Anthropic. Claude Code features described here follow Anthropic’s public documentation at the time of writing and change frequently; prices and model lists on this page are read from this gateway’s live settings.