How to Use Claude Skills in Claude Code and claude.ai

Claude skills are reusable instruction packs that teach Claude how to handle specialized jobs when your request matches the skill description. Each skill is a folder with a SKILL.md file at the root: YAML front matter plus markdown steps. You install skills on your machine or in your repo for Claude Code, or enable them in supported claude.ai settings when Anthropic exposes that path for your plan. This guide covers how to use claude skills from first folder to daily workflow: install paths, description triggers, testing, claude.ai versus Claude Code behavior, and drafting SKILL.md on PromptMake before you commit. PromptMake at https://promptmake.net/skills generates skill config text you copy into your project. It does not upload skills to Anthropic, create marketplace listings, or run Claude Code for you. This page is install and use, not marketplace browsing.

Before you install: what you are adding

A skill is not an app and not a plugin store purchase. It is markdown instructions Claude loads when the description trigger matches your task.

You need one primary job per skill folder. Release cuts, incident postmortems, OpenAPI style checks, and design review checklists belong in separate folders with separate description triggers.

Skills complement CLAUDE.md baseline repo notes. Put always-on stack facts in CLAUDE.md. Put conditional playbooks in skills so unrelated tasks stay lean.

As of mid-2026, Anthropic documents Agent Skills for Claude Code and expanding claude.ai support. Field names and UI labels change. Verify Anthropic help pages for your plan before you assume claude.ai paths match Claude Code exactly.

If you only need a one-time answer, skip skills and use chat. Skills pay off when the same guidance would be pasted every week.

Install claude skills for Claude Code

Claude Code reads skills from configured directories on your machine. Two common layers exist: personal skills in your user config path and project skills inside the repo your team clones.

Personal skills apply across projects. Use them for preferences that should not be committed: your private CLI aliases, your note-taking format, your local tooling wrappers.

Project skills travel with git. Use them for team law: release steps, migration fences, doc templates, security review checklists. Teammates inherit the same packs when they pull.

Installation steps in practice: create a folder named after the skill job, add SKILL.md with front matter and body, place the folder in the skills path Claude Code documents for your OS, restart or reload Claude Code if your version requires it, run a test prompt that should trigger the skill.

Optional sibling files sit beside SKILL.md: templates, example configs, small scripts, reference markdown. Link them from the body with relative paths.

Personal skills path

Anthropic documents a user-level skills directory per operating system in Claude Code setup guides. Create the directory if it does not exist. Drop one folder per skill inside it.

Keep personal skill names distinct from project skills to avoid confusion in logs. Prefix personal packs if your team uses shared naming conventions.

Do not commit secrets in personal skills. Reference environment variable names only.

Project skills path

Many teams use a .claude/skills or docs/claude-skills directory in the repo. Pick one path and document it in your onboarding README.

Commit SKILL.md with the code it affects. If release steps change, update the skill in the same pull request.

Add a PR checklist item: does this change need a SKILL.md update?

SKILL.md minimum viable install

Front matter requires at least name and description. Name is kebab-case: release-cut, openapi-style-check. Description is the trigger sentence.

Body holds purpose, when to use, numbered steps, fences, and examples. One happy path example and one missing-info example is enough for v1.

Draft structure at https://promptmake.net/skills or https://promptmake.net/claude-skill-generator if blank files stall you. Edit every command and path before commit.

Enable and use skills in claude.ai

claude.ai skills support evolves by plan and region. As of mid-2026, Anthropic rolls out Agent Skills beyond Claude Code for subscribers who opt in through settings or project features. Open Claude settings, look for Skills or Projects sections, and follow current help articles for upload or sync steps.

Browser workflow differs from terminal workflow. You may upload a skill folder, attach skills to a project, or select from skills already on your account depending on UI version. Claude Code instead reads local directories automatically when configured.

Description triggers work the same way in principle: Claude compares your prompt to each skill description and loads matches. Test with the same prompt list you use in Claude Code.

claude.ai sessions may not expose every Claude Code tool. Steps that say run npm test belong in coding skills used with Claude Code. Steps that say draft a customer email belong in claude.ai skills.

When Anthropic updates claude.ai skills UI, re-read release notes. This guide describes mid-2026 patterns, not a frozen screenshot.

Upload versus local sync

Some setups require zip upload of the skill folder through claude.ai settings. Some allow linking a repo path through desktop or connector products. Pick the path your plan documents.

After upload, run three test prompts: one that should load the skill, one that should not, one missing a required input. Log results the same way you log Claude Code tests.

claude.ai limits to plan for

Skills that reference local repo files only work in Claude Code with that repo checked out. claude.ai skills should link to content you upload or paste into project knowledge, not paths on your laptop.

Long SKILL.md files burn context when loaded. Move reference material to sibling files and link from the map section at the top of SKILL.md.

Do not treat claude.ai skills as a marketplace. You author or generate packs yourself. Public repos are inspiration, not one-click install from PromptMake.

How description triggers work when you use a skill

Claude evaluates available skills against your prompt. When the description semantically matches the task, the skill loads into context for that turn or session segment.

Triggers are soft matches, not regex gates. Write descriptions with verbs and artifacts your team types: deploy, migrate, lint, changelog, release, audit.

Over-triggering happens when descriptions are too broad: helps with code loads on every ask. Under-triggering happens when descriptions omit common phrases: a release skill that never says changelog may miss weekly release notes requests.

Update description before you rewrite the entire body when tests show misfires. Version SKILL.md in git like source code.

Test prompts that should load the skill

Build a list of five prompts that must activate the skill. Example for release-cut: cut the release branch for v2.4, prep this week's release PR, update changelog for tomorrow's ship.

Run the list after every description edit. Aim for zero false positives before you optimize false negatives.

Test prompts that should not load the skill

Build five adjacent tasks that should stay in generic Claude mode. Example for release-cut: fix typo in README, explain this function, write unit test for parser.

False positives waste context and can push wrong fences. Fix description scope before adding body length.

Missing-info behavior

Skills should ask for required inputs instead of guessing. A release skill asks for version number when absent. A migration skill refuses production when the user did not name staging.

Teach that behavior in Examples with a missing-info user ask and the expected clarification response.

Daily workflow: using claude skills on real tasks

Start the session with a normal natural-language ask. Do not memorize skill names as slash commands. Claude Code and claude.ai discover skills by description match.

When the skill loads, watch whether steps follow SKILL.md order. If the agent skips a fence, tighten wording in the body and rerun your test log.

When the skill does not load but should, copy the description trigger phrases into your ask once as a diagnostic. Then fix description permanently.

When two skills fight, narrow descriptions or merge skills. Split release cuts and hotfix recovery into separate folders.

Quarterly maintenance: run three prompts from each skill test log, fix stale commands, archive skills that reference removed scripts.

Claude Code session flow

Open repo with project skills committed. Confirm personal skills path if you use both layers. Ask for the job in plain language. Approve tool use per Claude Code rules when steps propose shell commands.

Log misfires in team wiki with prompt, expected skill, loaded yes or no, notes. Skills rot when descriptions change but tests do not rerun.

claude.ai session flow

Open project with skills attached if your plan supports projects. Ask for the job. Upload any reference files the skill links when those files are not in cloud knowledge.

For creative or policy skills without repo commands, claude.ai may be the better host. For npm test and git operations, prefer Claude Code.

Draft SKILL.md with PromptMake before install

Blank SKILL.md files stall teams. PromptMake https://promptmake.net/skills turns a rough job description into structured markdown: front matter, purpose, steps, fences, examples.

Guests get about three runs per day per path. Registered free users get about five. Quotas are separate from /loop-prompt-generator and /text paths.

Honest limits: the generator does not create folders on disk, does not register skills in Anthropic products, and does not run your test log. You copy output, place files, test locally, commit.

Practical path: write three sentences about the repeating job offline. Generate once. Replace placeholder commands with real scripts. Narrow description trigger. Run test prompts. Commit when false positives are zero.

For Custom GPT or Gemini Gem configs, use other tabs on /skills. This article stays on claude skills install and use.

Paste-ready fields from the generator

Expect name and description in YAML front matter, then body sections you can rename to match team style: Purpose, When to use, Steps, Fences, Examples.

Edit description until your five must-load prompts activate reliably. Edit steps until fences match your security policy.

Save winning output in your skill folder as SKILL.md. Add sibling reference files when the body gets long.

Common mistakes when using claude skills

Mistake 1: Installing from a marketplace listing without reading SKILL.md contents. You still must audit steps and fences.

Mistake 2: Generic description triggers. Fix description before adding body length.

Mistake 3: One mega-skill for unrelated jobs. Split by outcome.

Mistake 4: Expecting PromptMake to install the skill because markdown was generated. You still copy files and test.

Mistake 5: Duplicating conflicting guidance in CLAUDE.md and three skills.

Mistake 6: Steps that reference files missing from the skill folder.

Mistake 7: No fences around destructive shell commands.

Mistake 8: Never rerunning tests after a description edit.

Mistake 9: Using claude.ai skills with repo-only file paths that do not exist in the browser session.

Skills versus other instruction layers

CLAUDE.md sets baseline repo context. Skills load conditionally. Cursor rules and editor hooks may overlap. Pick one source of truth per workflow.

Loop commands in Claude Code are session-scoped with verify and turn caps. Skills are reusable packs. Use loops for one ticket with a finish line. Use skills for standing playbooks.

Custom GPTs and Gemini Gems solve similar reuse problems on other platforms. PromptMake /skills cross-links formats when you standardize jobs across vendors.

Soft next steps

Pick one weekly chore you repeat in Claude. Draft three sentences offline. Generate SKILL.md at https://promptmake.net/skills. Install in Claude Code project path or enable in claude.ai per your plan docs. Run five must-load and five must-not-load prompts. Commit when false positives are zero.

Read claude-code-skills-guide on this blog for deeper file layout and test loops. Read what-are-claude-skills for the builder intro. This page stays on how to use claude skills day to day.

FAQ

How do I use claude skills in Claude Code?

Create a folder with SKILL.md at the root. Place it in the personal or project skills directory Claude Code documents for your OS. Write a specific description trigger in front matter. Ask for work in natural language. Claude loads matching skills automatically. Test with prompts that should and should not activate the skill.

Can I use claude skills in claude.ai?

As of mid-2026, Anthropic expands skills support in claude.ai by plan. Open settings or project features and follow current help articles for upload or attach steps. Browser skills may not run local repo commands. Match skill steps to the host you use.

How do I install a claude skill from scratch?

Create a folder named for the job. Add SKILL.md with YAML front matter containing name and description, plus markdown steps and examples. Copy the folder to your Claude Code skills path or upload through claude.ai per your plan. Run test prompts before you rely on the skill in production.

Does PromptMake install claude skills for me?

No. PromptMake https://promptmake.net/skills generates SKILL.md text you copy into your folder. You place files, test triggers, and commit. PromptMake does not upload to Anthropic or operate a skills marketplace.

What is the description trigger in SKILL.md?

The description field in YAML front matter tells Claude when to load the skill. Write one or two sentences with verbs and scope. Claude compares your prompt to descriptions semantically. Precise descriptions reduce wrong loads.

How is this different from the claude skills marketplace?

Marketplace and public repo posts cover what others publish. This guide covers how to use claude skills you author or generate: install paths, triggers, testing, Claude Code versus claude.ai. PromptMake generates config text; it does not sell or host third-party skill listings.

Should I use a skill or paste instructions each chat?

Paste instructions for one-off tasks. Use a skill when the same guidance repeats weekly and you want conditional loading without bloating every session. Skills pay off when triggers are tuned and tests stay green.

How do I try PromptMake skills generation free?

Open https://promptmake.net/skills. Guests get about three generations per day per path without signup. Registered free accounts get about five per day. Generate SKILL.md, edit paths and fences, install locally, run your test prompt list.