← Back to library

Anthropic Agent Skills

Self-contained instruction folders teach Claude repeatable workflows across documents, development, design, and enterprise tasks.

What problem it solves

Anthropic Agent Skills is a set of task instructions that Claude reads on demand. Each skill sits in its own directory, using SKILL.md to hold metadata, a trigger description, and operating instructions, and it can also come with scripts and resources. Users do not have to repeat rules, steps, and delivery formats in every conversation; they only need to select or name the relevant skill.

A typical run goes like this: submit the task, the input files, and the necessary constraints, and Claude then loads the instructions, scripts, and resources of the matching skill to produce documents, content analysis, or other artifacts. What sets it apart from writing a one-off long prompt is that the workflow can be saved, reused, and reviewed per directory, and it fits a team's fixed way of doing things better.

How to use it

In Claude Code, first register the repository as a plugin marketplace, then install the document skills or the example skills:

text
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills

Once installed, simply name the relevant skill in your request to trigger it. Paid Claude.ai plans already include these example skills; the Claude API also supports calling preset skills or uploading custom skills.

Inputs and results

Inputs usually include the task description, the files to process and their formats, and constraints such as permissions. Claude runs the specified workflow according to the chosen skill and outputs documents, analysis results, or other task artifacts. The actual read and write scope is determined by the skill's contents; a PDF skill, for example, can read a specified file and extract form fields. Before putting it to real use, check SKILL.md, the bundled scripts, and the file permissions the skill can access.

Who it is for

It suits people who want to codify team workflows for Claude to execute, and also developers studying the directory structure of Agent Skills. The repository provides sample skills, a canonical directory layout, and basic templates that help you understand how a skill organizes files, describes trigger conditions, and writes operating instructions.

Usage boundaries

This repository is mainly for demonstration and teaching. Anthropic recommends testing it yourself before critical tasks, since Claude's actual behavior may differ from the samples. Most skills use Apache 2.0, but document skills such as docx, pdf, pptx, and xlsx are source-available only, not open source; the example skills in Claude.ai are limited to paid plans.