The pro reference for working with Claude across every surface — skills, MCP servers, plugins, slash commands, hooks, the API, and the 4.x model family. Built for engineers who ship.
A portable, model-agnostic capability bundle. Three pieces, one folder.
YAML frontmatter says when the skill loads. Markdown body says what to do. That's the whole interface — no SDK, no compile step, no runtime.
Scripts, templates, schemas, reference docs — anything Claude reads or executes while the skill is active. Loaded lazily.
Only the description is always loaded. Body + bundled files pulled only when the skill triggers.
Same folder works in Claude Code, API, claude.ai, and desktop.
allowed-tools caps what a skill can do. Safer by default.
Skills stack. One prompt can light up many at once.
A short tour from the official Anthropic catalog and the community.
Skills work in Claude Code (CLI + IDE), the Claude API, claude.ai, and the desktop apps.
Drop a skill folder into ~/.claude/skills/ (user) or .claude/skills/ (project). Restart the session — Claude auto-discovers it.
# user-wide
git clone https://github.com/anthropics/skills ~/.claude/skills-src
cp -r ~/.claude/skills-src/document-skills/pdf ~/.claude/skills/
Full Claude Code instructions →
Use the code_execution tool plus the Skills container. Upload skill folders to the Files API and reference them in your request.
# Anthropic SDK client.beta.messages.create( model="claude-opus-4-7", betas=["skills-2025-10-02"], container={"skills": ["pdf", "xlsx"]}, ... )Full API instructions →
Settings → Capabilities → Skills. Toggle the official Anthropic skills (PDF, Word, Excel, PowerPoint) or upload a custom one as a folder.
Walkthrough with screenshots →Some skills ship as part of a Claude Code plugin or MCP server (e.g., Figma, Playwright). Install via the plugin marketplace or claude plugin install.
Short answers to the questions people ask most often about Claude, skills, MCP, and the API.
A skill is a folder containing a SKILL.md file (YAML frontmatter + markdown body) plus optional scripts and reference files. Claude loads the skill on demand when a user's prompt matches its description — extending what Claude can do without bloating every prompt. See the concept guide.
User-scope skills go in ~/.claude/skills/<skill-name>/. Project-scope skills go in .claude/skills/ inside the repo. Claude Code auto-discovers both at session start. For claude.ai, upload skills via Settings → Capabilities → Skills.
MCP is the open JSON-RPC protocol Claude uses to talk to external systems — databases, APIs, your own tools. You install MCP servers (Postgres, GitHub, Slack, Figma, etc.) and Claude can call their tools mid-conversation. Read the MCP guide.
Default to Claude Sonnet 4.6 — strong, fast, ~5× cheaper than Opus. Use Claude Opus 4.7 for hard reasoning, large refactors, and multi-step plans. Use Claude Haiku 4.5 for high-volume routing and classification. See the model comparison.
Run npm install -g @anthropic-ai/claude-code, then claude in your project. Sign in with claude.ai (subscription) or an API key. Full setup walkthrough in the Quickstart.
Skills trigger automatically from natural-language matching. Slash commands are user-typed shortcuts (/review). MCP servers are long-running processes that expose tools Claude can call. They layer together — skills can call MCP tools, slash commands can invoke skills.
Mark a stable prompt prefix (system prompt, big context doc, tool definitions) with cache_control: {"type": "ephemeral"} and cache hits within 5 minutes are ~90% cheaper. Heavy users see the cache extend for hours. See the API caching guide.
No. ClaudeBro is an unofficial community resource. All Claude trademarks and skill names belong to their respective owners. Official documentation lives at docs.claude.com.