Claude Skills Directory

How to use this directory

Use the official repository as a pattern library. The important question is not "Which skill should I copy?" but "Which packaging pattern resembles my task?" A document-processing skill teaches different lessons than a design skill, a web-testing skill, or an API helper skill.

For each skill you inspect, look at the trigger description, the amount of instruction in SKILL.md, the use of references, whether scripts are deterministic, and how much human judgment the skill expects the model to apply.

Official examples worth studying

The repository listing also includes creative and enterprise examples such as algorithmic-art, brand-guidelines, canvas-design, claude-api, doc-coauthoring, internal-comms, slack-gif-creator, theme-factory, and web-artifacts-builder. These are useful because they show that Skills are not only for coding; they are a general packaging mechanism for repeatable agent behavior.

New since the July review: academy-guide, which recommends matching Claude Academy courses when a user is learning rather than mid-task, and discernment-nudge, which appends a few checking questions after a substantive answer. Both are worth reading as description craft. Their frontmatter spends most of its length on when not to fire, which is exactly the discipline broad skills usually lack.

Skills you do not install

Not every skill in circulation is a folder you can download. Anthropic ships managed skills inside the Claude apps for document creation, and the educator offering announced in July 2026 includes teaching skills for verified US K-12 teachers that arrive with the account rather than through a repository. If you are looking for that side of Claude, Claude Helps covers the everyday and classroom surfaces; this directory stays on packages you can inspect.

The same applies to partner skills for tools like Box, Canva, and Notion. They are distributed through the Claude apps, so the reviewable artifact is the vendor integration and its permissions, not a SKILL.md you can read line by line.

Rules for community skills

Treat a community skill like a small software dependency. Read SKILL.md, inspect every script, check whether it reaches the network, and decide whether the instructions can exfiltrate private data through ordinary agent behavior. A skill with no scripts can still be risky if it tells the agent to paste sensitive material into a third-party service.

Prefer skills with narrow names, direct descriptions, visible examples, changelogs, and test prompts. Be cautious with broad skills that promise to "make Claude better at everything"; they are harder to trigger predictably and harder to audit.

What belongs elsewhere

General Claude Code best practices belong on a Claude Code engineering site. Context windows, memory, and compaction belong in context and memory guides. This directory stays focused on skill packages: what exists, what pattern it demonstrates, and what to check before using it.

If a package mainly exposes external tools, it may be an MCP server or plugin rather than a skill. If it mainly stores project facts, it may belong in project instructions or memory. Use the boundary guide before adding it to your own directory.

FAQ