Anthropic Official Skills: Complete Guide to 19 Open-Source Agent Skills

Updated October 21, 2026: The repository now holds 19 skills — academy-guide and discernment-nudge joined in August 2026 — and this guide adds them. Also refreshed: the claude-api model table (Opus 5.5 is now the default), the effort levels, the frontend-design guidance, the PowerPoint font rules, and the marketplace plugin list.

Anthropic has open-sourced 19 Agent Skills at anthropics/skills, covering creative design, document creation, technical development, enterprise communication, and response guidance. These skills work across Claude.ai, Claude Code, and the Claude API. This guide provides a deep dive into every skill.

What Are Agent Skills?

Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. Each skill’s core is a SKILL.md file with YAML frontmatter and markdown instructions.

---
name: my-skill-name
description: What this skill does and when to use it
---

# My Skill Name
[Instructions Claude will follow]

Installation

Claude Code:

/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills
# Also available as their own plugins:
/plugin install claude-api@anthropic-agent-skills
/plugin install academy-guide@anthropic-agent-skills
/plugin install discernment-nudge@anthropic-agent-skills

Claude.ai: the example skills are already available on paid plans.

Claude API: See the Skills API Quickstart.

Skill Loading (Progressive Disclosure)

Metadata — Always visible (~100 word description)
SKILL.md — Loaded on trigger (<500 lines)
References — Loaded on demand (scripts, data, templates)

All 19 Skills at a Glance

Category	Skill	Purpose
Creative	algorithmic-art	p5.js generative art with seeded randomness
	canvas-design	Museum/magazine-quality visual art
	slack-gif-creator	Slack-optimized animated GIFs
	theme-factory	10 professional color themes
Design	brand-guidelines	Anthropic brand identity
	frontend-design	Production-grade UI, anti-”AI slop”
	web-artifacts-builder	React + Tailwind interactive artifacts
Documents	docx	Word document creation/editing
	pdf	PDF read/create/merge/OCR/forms
	pptx	PowerPoint presentation creation
	xlsx	Excel spreadsheets with formulas
Technical	claude-api	Claude API/SDK development
	mcp-builder	MCP Server construction
	skill-creator	Skill creation with eval testing
	webapp-testing	Playwright browser automation
Enterprise	doc-coauthoring	Collaborative document writing
	internal-comms	Internal communication templates
Guidance	academy-guide	Suggest matching Claude Academy courses and tutorials
	discernment-nudge	Append follow-up questions that help you check an answer

Creative & Generative Skills

Algorithmic Art

Creates generative art using p5.js with seeded randomness and interactive parameter exploration.

Workflow:

Create Algorithmic Philosophy (.md) — define aesthetic direction
Implement p5.js interactive piece (.html) — using provided viewer template

Features:

Seeded randomness for reproducibility (Art Blocks pattern)
Interactive parameter controls and seed navigation
Particle systems, noise fields, flow fields
Philosophy: beauty in process, not final frame

Canvas Design

Museum/magazine-quality visual art in .png and .pdf.

90% visual, 10% text — text only as visual accent
Sophisticated design-forward approach (never cartoony)
Bundled canvas-fonts directory
Emphasis: appears hand-made, not AI-generated

Slack GIF Creator

Animated GIFs optimized for Slack.

Spec	Value
Emoji size	128x128
Message size	480x480
FPS	10-30
Colors	48-128
Duration	<3 seconds (emoji)

Built-in: GIFBuilder class, 7 easing functions (linear, ease_in, ease_out, ease_in_out, bounce_out, elastic_out, back_out), 8 animation concepts (shake, pulse, bounce, spin, fade, slide, zoom, particle burst).

Theme Factory

10 pre-set professional themes with colors and fonts.

Themes: Ocean Depths, Sunset Boulevard, Forest Canopy, Modern Minimalist, Golden Hour, Arctic Frost, Desert Rose, Tech Innovation, Botanical Garden, Midnight Galaxy.

Each theme: complete color palette (hex codes) + complementary font pairings. Applicable to slides, docs, reports, landing pages.

Design & Frontend Skills

Brand Guidelines

Applies Anthropic’s official brand identity. Colors: Dark #141413, Light #faf9f5, Orange #d97757, Blue #6a9bcc, Green #788c5d. Fonts: Poppins (headings), Lora (body).

Frontend Design

Production-grade frontend interfaces that avoid “AI slop” aesthetics.

Principles:

Ground the design in the subject matter before choosing anything visual
Deliberate typefaces and a real type scale; type treatment as part of the design
Structure that encodes information (numbered markers only for actual sequences)
Motion used sparingly: one orchestrated moment beats scattered effects
Spend your boldness in one place, then cut one accessory before shipping

Calibrated against today’s AI-design tells: cream background plus high-contrast serif and a terracotta accent, near-black with one acid accent, broadsheet hairline columns, the SaaS-card kit (identical rounded cards, one radius, the same soft grey shadow), and template chrome such as ALL-CAPS eyebrows, middle-dot meta strings, and a → tacked onto every link.

Web Artifacts Builder

Multi-component interactive artifacts using React 18 + TypeScript + Tailwind + shadcn/ui.

40+ pre-installed shadcn/ui components + Radix UI
scripts/init-artifact.sh — project initialization
scripts/bundle-artifact.sh — bundle to single self-contained HTML
Parcel bundler with path alias configuration

Document Skills (Source Available)

These four skills power Claude’s document capabilities. Source-available (not open source) for developer reference.

DOCX — Word Documents

Operation	Tool
Read	pandoc (text), unpack XML (raw)
Create	docx-js library
Edit	Unpack -> edit XML -> repack

Supports: tables (DXA units), images, headers/footers, tracked changes, TOC, multi-column layouts, hyperlinks, bookmarks, footnotes.

PDF — PDF Processing

Operation	Tool
Read/Extract text	pdfplumber
Extract tables	pdfplumber + pandas
Create	reportlab (Canvas or Platypus)
Merge/Split/Rotate	pypdf
OCR	pytesseract
Forms	Detailed in FORMS.md
CLI operations	qpdf

PPTX — PowerPoint

Operation	Tool
Read	markitdown (text), thumbnail.py (visual)
Create	pptxgenjs (npm)
Edit	Unpack -> edit XML -> clean -> repack

Design rules: bold, content-informed palettes with one color at 60-70% visual weight, titles 36-44pt, body 14-16pt, no accent lines under titles and no decorative color bars. Fonts are a QA question, not a taste one: Arial, Calibri, Cambria, Times New Roman, Courier New, Bookman Old Style and Century Schoolbook render true-to-width in the skill’s LibreOffice check, while Georgia, Trebuchet MS, Impact, Garamond and friends substitute at different widths, so text-fit checks on them can’t be trusted. Visual QA is mandatory — assume problems exist.

XLSX — Excel Spreadsheets

Tools: pandas (data analysis), openpyxl (complex formatting/formulas).

Critical rules:

Use formulas (=SUM()), NEVER hardcoded values
Run scripts/recalc.py after formula creation
Financial model color coding: Blue (inputs), Black (formulas), Green (internal links), Red (external links)
Validate zero formula errors (#REF!, #DIV/0!, #VALUE!, #N/A, #NAME?)

Technical & Development Skills

Claude API

Build LLM applications using Claude API or Anthropic SDKs.

Models (cached in the skill as of October 2026):

Model	Model ID	Use
Fable 5.1	`claude-fable-5-1`	Most capable widely released model
Opus 5.5	`claude-opus-5-5`	The skill’s default — use it unless the user names another
Sonnet 5.5	`claude-sonnet-5-5`	Speed and capability for everyday coding
Haiku 5.5	`claude-haiku-5-5`	Fast, low cost

Key features:

Thinking: thinking: {type: "adaptive"}
Effort: output_config: {effort: "low"|"medium"|"high"|"xhigh"|"max"}
Streaming: default for long input/output
Tool Use: automatic runner or manual loop
Structured Outputs: output_config: {format: {...}}
Files API: persistent file references
SDKs: Python, TypeScript, Java, Go, Ruby, C#, PHP, plus raw cURL — every one supports the beta Tool Runner and Managed Agents

Decision tree: Single call -> Workflow (tool use) -> Agent (open-ended exploration).

MCP Builder

Create high-quality MCP (Model Context Protocol) servers.

Four phases:

Deep Research — API docs and integration patterns
Implementation — TypeScript recommended, Python alternative
Review/Test — MCP Inspector (npx @modelcontextprotocol/inspector)
Evaluations — 10 complex, independent evaluation questions

Principles: Balance comprehensive API coverage with workflow tools, and default to comprehensive coverage when unsure. Clear naming (github_create_issue). Actionable error messages. Tool annotations: readOnlyHint, destructiveHint, idempotentHint.

Skill Creator

Create new skills, iterate, and measure performance.

Workflow:

Draft SKILL.md + resources
Set up evals/evals.json (2-3 realistic prompts)
Evaluate: with-skill vs. baseline in parallel
Improve based on feedback
Expand test coverage
Optimize description triggering

Tools: eval-viewer/generate_review.py, run_loop.py (20-query description optimization), benchmarking (mean +/- stddev).

Webapp Testing

Test local web applications using Playwright.

Pattern: Reconnaissance-then-action (wait networkidle -> screenshot -> identify selectors -> execute)
scripts/with_server.py manages server lifecycle
Supports multiple servers simultaneously (backend + frontend)
Always wait for page.wait_for_load_state('networkidle') before DOM inspection

Enterprise & Communication Skills

Doc Coauthoring

Structured workflow for collaborative document creation.

Stage	Process
1. Context Gathering	User provides background, Claude asks 5-10 clarifying questions
2. Refinement	Section-by-section: brainstorm (5-20 options) -> curate -> draft -> iterate
3. Reader Testing	Fresh Claude (no context) validates document clarity

Quality gate: after 3 iterations with no changes, ask what can be removed.

Internal Comms

Company-specific internal communications using templates.

Types: 3P updates (Progress/Plans/Problems), company newsletters, FAQ responses, status reports, leadership updates, project updates, incident reports.

Each type has corresponding guidelines and templates in the examples/ directory.

Guidance Skills

These two skills shape how Claude answers rather than what it builds. Both are composable: they run alongside whatever other skill is handling the task.

Academy Guide

Checks the Claude Academy catalog when you ask how to use Claude or a Claude product, and mentions a matching course, tutorial, use case, or product hub at the end of a normal answer.

Rules: answer the question first, recommend only on a strong match of intent (someone learning a feature, not someone mid-task), and never invent Academy content.

Discernment Nudge

After a substantive answer you might act on — advice, a drafted plan or pitch, an estimate, an interpretation of data — appends two or three short follow-up questions tied to specifics in what it just produced, at most once per conversation.

The three habits it models: checking facts, questioning reasoning, and noticing missing context. It skips trivial lookups, pure explanations, formatting jobs, code you’re about to run, creative writing, and any request where you already asked Claude to double-check itself.

Cross-Cutting Design Patterns

Anti-AI-Slop — All creative skills emphasize distinctive, context-specific design over generic templates
Philosophy Before Implementation — Define aesthetic direction before coding
Formulas Over Calculations — Dynamic documents, not hardcoded values
Mandatory Validation — Visual QA for PPTX, zero errors for XLSX, evals for skills, 10 questions for MCP
Progressive Disclosure — Metadata always visible, details loaded on demand

Building Your Own Skill

my-skill/
├── SKILL.md           # Required: instructions + frontmatter
├── references/        # Optional: reference docs
├── scripts/           # Optional: helper scripts
├── assets/            # Optional: resources
└── evals/             # Recommended: evaluation tests
    └── evals.json

Every frontmatter field is optional, but description is what tells Claude when to use the skill. Uploads to claude.ai and the Skills API accept only the spec’s six fields — name, description, license, compatibility, metadata, allowed-tools — and fail outright on anything else, while Claude Code additionally accepts its own fields such as disable-model-invocation and allowed-tools grants. The markdown content contains instructions, examples, and guidelines.

Resources: