Tech & AI, Explained

claude-skills

7 min read

🇺🇸 USA · Claude Skills

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Reusable instruction packs that customise Claude’s behaviour for specific kinds of work. Like saved recipes — write once, reuse forever, share with team or community.

Front-matter facts

FieldValue
VendorAnthropic (San Francisco, USA)
Country / origin🇺🇸 USA
Recommended for Australian users?✅ Yes — same Claude infrastructure
Privacy summarySame posture as Claude Code overall; Skills themselves are local files unless you publish them
Free tierSkills are included with Claude Code (which is bundled with Claude Pro and above)
Paid tiersSkills work at all Claude Code tiers; some advanced Skills features may be Pro+
First released2025 — Skills system rolled out in Claude Code; expanded into claude.ai consumer over 2025-26
Last reviewed2026-06-26
Official sitehttps://docs.claude.com/en/docs/claude-code/skills

What it is

Claude Skills are reusable instruction packs for Claude. A Skill is essentially a folder containing:

  • A SKILL.md file — markdown instructions telling Claude how to do a specific kind of work
  • Resources — supporting files (templates, examples, reference docs, scripts) Claude can read
  • Tools — optional executable scripts the Skill calls

You invoke a Skill by name in Claude Code (or in claude.ai). Claude reads the SKILL.md, follows its instructions, uses the resources, and produces output tailored to that specific kind of work.

Examples of Skills:

  • encyclopedia-add — a Skill that knows how to add an entry to George’s personal encyclopedia following its conventions (this very encyclopedia uses one)
  • verify — a Skill that runs and tests a code change to verify it works
  • code-review — a Skill that reviews a diff for bugs and cleanup opportunities
  • security-review — a Skill that audits code for security issues
  • canvas-design — a Skill that creates poster / design artifacts
  • docx — a Skill that produces / edits Word documents
  • pptx — a Skill that produces / edits PowerPoint slides
  • brand-guidelines — a Skill that applies Anthropic’s brand styling

Skills are powerful because they encapsulate workflow + context + tools in a single addressable name. Instead of pasting the same long prompt every time you want to do X, you create a Skill once and invoke it forever.

Analogy: if Claude is a smart colleague, Skills are the binders of standard operating procedures on the shelf. “Do the quarterly report” doesn’t need you to re-explain how the quarterly report is structured every time — there’s a binder for that. Claude opens the binder, follows the instructions, asks for input only where needed.

What you’d use it for

Personal productivity Skills

  • Custom workflows you do repeatedly (“write a weekly update,” “draft a proposal,” “summarise an article”)
  • Per-project instructions (“when working in repo X, always run npm run lint before commit”)

Coding Skills

  • Project-specific patterns (“add a new API endpoint following our project’s conventions”)
  • Framework-specific guidance (“create a new Next.js page with our Tailwind + shadcn setup”)
  • Test generation matching your team’s style

Business / domain Skills

  • Brand-styled documents (matching company colours, fonts, voice)
  • Industry-specific compliance (“draft a privacy notice meeting Australian Privacy Act requirements”)
  • Repetitive multi-step research workflows

Anthropic’s bundled Skills

  • Anthropic ships a growing library of official Skills: docx, pdf, pptx, xlsx (Office documents), skill-creator (a Skill that helps you write new Skills), brand-guidelines, canvas-design, internal-comms, and more.

How to use Skills

Invoke an existing Skill

  • In Claude Code, type /skill-name or describe a task that matches a Skill’s description; Claude proposes invoking it
  • The Skills system intelligently picks Skills based on triggers in their descriptions

Create your own Skill

  1. Use the bundled skill-creator Skill: claude then ask “Create a new skill for X” — skill-creator walks you through it
  2. Or manually: create a folder ~/.claude/skills/my-skill-name/SKILL.md with the instructions
  3. Test it by invoking /my-skill-name or asking Claude to use it

Share / discover Skills

  • Anthropic Skills Marketplace (rolling out) — browse and install community Skills
  • GitHub — Skill folders are just files; share via git, install by cloning

Skill file structure

~/.claude/skills/encyclopedia-add/
├── SKILL.md             # The instructions (markdown frontmatter + body)
├── resources/           # Supporting reference docs
│   ├── conventions.md
│   └── style-guide.md
└── scripts/             # Optional executable helpers
    └── validate-entry.sh

SKILL.md frontmatter format

---
name: encyclopedia-add
description: Add a new entry to George's personal encyclopedia following its conventions. Trigger on /encyclopedia-add or "add to encyclopedia."
allowed-tools: Read, Write, Edit, Grep, Glob
---
 
# Instructions for Claude
 
[The actual instructions for the Skill go here in markdown...]

How it compares to alternatives

CapabilityClaude SkillsOpenAI GPTsGoogle Gemini GemsCursor / Windsurf Rules
FormatMarkdown SKILL.md + resourcesConfigured via UIConfigured via UIMarkdown rules files
Local / version-controlledYes (files)No (cloud-managed)No (cloud-managed)Yes (files)
Bundled with tools / scriptsYesLimited (Actions)LimitedNo
Shared via marketplaceYes (Anthropic marketplace)GPT Store (mature)Gems galleryNo (rules are private)
Auto-trigger on description matchYesNo (must invoke)NoAlways-on (per project)
Works in CLI agentYes (Claude Code)No (ChatGPT only)No (Gemini only)Yes
Works in chat surfaceYes (claude.ai rolling)YesYesNo
CostBundled with Claude ProNeed ChatGPT Plus to buildNeed Google AI ProBundled with Cursor / Windsurf

Skills are particularly powerful for repeated workflows you do in a CLI agent context — claude.ai’s web GPTs equivalent is good for chat-flavoured custom assistants, but Claude Skills shine for “I do this code-shaped or doc-shaped task often, save the recipe.”

Privacy / data handling

  • Skills you create locally are files on your disk — never sent to Anthropic unless you choose to share or publish them
  • When a Skill runs, the SKILL.md content + any resources Claude reads are sent as context to the Claude model (same as any other Claude conversation)
  • Skills published to the Anthropic Skills Marketplace are public
  • For confidential workflow patterns, keep Skills local and version them in private git

Recent changes

  • 2026: Skills system matured; Claude Skills Marketplace launched
  • 2026: Claude.ai consumer chat gained Skills support
  • 2025: Skills launched in Claude Code; rapidly adopted
  • 2025: Anthropic shipped many bundled Skills (docx, pptx, xlsx, pdf, skill-creator, etc.)

Gotchas

  • Skills must be discoverable by Claude to be auto-triggered — the description field in SKILL.md frontmatter is what Claude matches user queries against. Write descriptions that include trigger words / phrases.
  • Avoid Skill bloat — too many narrow Skills means Claude has to pick from many similar options. Prefer fewer, broader Skills.
  • Skills can call tools but the tools must be granted by the user — allowed-tools frontmatter restricts what a Skill can do.
  • Resources are NOT auto-loaded — the SKILL.md must explicitly tell Claude to read specific resource files when relevant. Otherwise resources sit unused.
  • Skills can be confused with Plugins — Plugins (Claude Code) are bundles of Skills + Commands + MCP connectors + Hooks; Skills are one specific format within that.
  • The encyclopedia uses a Skill (encyclopedia-add) — this very file follows conventions documented in that Skill. Local Skills make personal projects significantly faster.
  • Sensitive Skills shouldn’t go in the public marketplace — once published, they’re public.
  • Skills work across Claude Code AND claude.ai but the experience differs slightly — Claude Code’s auto-discovery is more refined as of mid-2026.

See also

Sources

Graph View

Backlinks