Claude Code Skills: install an existing skill and create your own

If every form review starts with you pasting the same checklist into Claude Code, that checklist probably belongs in a skill. Write down the steps, the examples and the report you expect once, and reuse them next time. The same goes for drafting change descriptions or working through build errors.

A Skill is a folder containing instructions in SKILL.md and, optionally, supporting material. It hands the agent a repeatable procedure together with its examples and helper files. The format is described in the Agent Skills specification.

• Agent Skills specification

This guide covers installation in local Claude Code and walks through a sample skill that reviews an enquiry form. Commands were checked against the documentation on October 3, 2026. The example is for teaching and is not a report of testing on a live website.

Choose a task for a Skill

Pick a procedure you can already explain in words: “Before handing over a form, inspect required fields, error messages and submission behavior, then report problems.”

That makes a better first skill than “build quality websites.” You can name the files to inspect, the checks to make and the findings you expect back.

Start with one row of the table. A skill that builds a site, writes the ads, configures the server and publishes the lot is hard to fix, because you cannot tell which stage went wrong.

Install a skill from a folder

Personal skills live in `~/.claude/skills/`, project skills in `.claude/skills/` inside the repository. Each skill gets its own folder with a SKILL.md. A personal skill is available in every project on that machine, while a project skill travels with the code.

• Skill locations

project/
└── .claude/
    └── skills/
        └── form-review/
            ├── SKILL.md
            └── references/
                └── checklist.md

If you downloaded a collection, find the skill you want and copy its entire folder. Copying SKILL.md on its own can break references to the supporting files.

Unpacking an archive often adds an extra level of nesting, so check the resulting tree:

.claude/skills/form-review/form-review/SKILL.md

In our example the file should sit one level higher:

.claude/skills/form-review/SKILL.md

Before copying, check whether a folder with that name already exists. If it does, compare the versions first, or you may overwrite your own changes.

Install skills through a plugin

If the author ships skills as a plugin, follow their installation instructions. For the official Anthropic repository, type these commands inside Claude Code:

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

The first command adds the marketplace and the second installs the example skills. For document work, the same repository lists `document-skills@anthropic-agent-skills`.

• anthropics/skills README

A folder and a plugin are different ways of distributing a skill. A lone SKILL.md you found on GitHub cannot be installed with `/plugin install`. Check first whether the author offers a marketplace package.

`/plugin install` opens the plugin's details, where you choose the scope: for you everywhere, for everyone working in the repository, or for you in this repository only. Read the summary Claude Code prints afterwards. If a reload is needed, Claude Code normally runs `/reload-plugins` for you. If the plugin failed to load, the reason is on the Errors tab in `/plugin`.

• Plugin installation

Inspect someone else's skill

Read SKILL.md and any bundled scripts before the first run, and compare what they do with the task. A skill meant to review a form that also sends your repository to an external server owes you an explanation.

Note where the skill came from, which version you are using and what you changed. If an update alters the results, you will have something to compare against.

Look at the assumptions as well. The author may have written the skill for a particular framework, command set or browser tool that your project does not have. Replace those parts with your own. “Works everywhere” is not something a skill can promise.

Create a form-review Skill

Our skill reads the form's implementation and writes a report. It does not fix code or submit enquiries, which keeps finding a problem separate from repairing it.

Create the `.claude/skills/form-review/` folder in your editor. On macOS, Linux or WSL you can use the terminal:

mkdir -p .claude/skills/form-review

Then create SKILL.md inside it with the content below. It is a teaching example, so adapt the procedure to your project.

---
name: form-review
description: Reviews an enquiry form implementation and reports on validation, messages and submission. Use for form review before release.
disable-model-invocation: true
---

# Enquiry form review

## Goal
Find implementation problems and identify remaining checks.
Read files only. Do not change code, submit the form
or contact external services.

## Inputs
A form path or user journey name is required.
If several forms match, ask which one to review.

## Procedure
1. Read the form component, handler and related tests.
2. Find required fields and validation rules.
3. Inspect handling of empty input and invalid email in the code.
4. Trace submitting, success and error states.
5. Check that repeated clicks during submission are handled explicitly;
   if unclear, describe the uncertainty.
6. Compare findings with tests. A test's existence does not mean
   it was executed or passed.

## Report
For each issue, provide:
- file and code location;
- conditions that trigger it;
- possible user impact;
- suggested verification or correction.

List separately behavior that requires running the application.
If no clear issues were found, say so without claiming
the form was fully tested.

The metadata sits at the top, between the `---` lines, and the procedure follows. The open Agent Skills specification requires `name` and `description`, and the name must match the folder. Claude Code itself is more lenient: every field is optional and only `description` is recommended. Names may use lowercase Latin letters, digits and hyphens.

• SKILL.md format

`disable-model-invocation: true` is a Claude Code setting. With it, the skill runs only when you call it, and Claude will not load it on its own. We use it here so that a review always starts with an explicit command. The words “read files only” in the instructions describe behavior. Technical permissions are configured separately.

• Invocation control
• permissions

Invoke it and check the result

Open the project in Claude Code and type:

/form-review Review the enquiry form in src/components/LeadForm.tsx

The path is only an illustration, so point it at a real file in your project. A skill is invoked directly as `/skill-name`.

• Invoking Skills

To test the skill, choose a form with a quirk you already know about. Say the submission error disappears the moment the button is clicked again. See whether the agent spots exactly that and can show the cause in the code.

Judge the report on four points:

1. It cites files that exist and specific places in the implementation.
2. Every finding states the condition that triggers the problem.
3. Assumptions are kept apart from conclusions the code supports.
4. Having read the tests is not passed off as having run them.

If you get a long list of generic advice instead, tighten the evidence requirement. For example, add: “Do not include a recommendation without identifying the relevant implementation.” Then run the review again on the same form.

Change one substantial instruction at a time, otherwise you will not know which edit made the answer more useful. This describes an editorial method. We are not reporting skill-test results here.

Write a description for automatic selection

If Claude is meant to pick the skill automatically, describe both what it does and when to use it. Compare:

Helps develop applications.

And a more precise version:

Reviews enquiry forms before release: field validation, error messages and submission states. Use for form-review requests, not general architecture reviews.

The second version shows its boundaries, which gives the agent fewer reasons to pick the skill for an unrelated task. The specification also recommends that `description` say what the skill does and when to use it.

• Description requirements

Automatic selection is switched off in our example. If you turn it on, test the description with several requests: one that fits, one on a neighboring topic and one that is clearly unrelated. Watching what happens tells you more than judging the wording by ear.

If the Skill is missing or does not trigger

Start with the path, the exact SKILL.md filename and the top of the YAML block: the opening `---` must be the first line. Claude Code picks up edits inside existing skill folders during a session. If you created the top-level `.claude/skills/` directory after the session started, run `/reload-skills`.

• Editing and loading

Then separate two situations. If the skill is not found when you call it directly, check the installation. If it runs manually but the agent never picks it on its own, check the invocation mode and the description.

When a personal and a project skill share a name, Claude Code runs the personal one. If the result surprises you, look in both folders.

• Name collisions