Skip to main content
Claude Skills is a library of structured workflows — called skills — that you invoke with a single slash command in Claude Code. Each skill encodes domain knowledge, step-by-step processes, quality gates, and self-review checklists so Claude handles complex tasks reliably, every time.
Skills live in ~/.claude/skills/ and are available in every Claude Code session automatically.

The problem skills solve

Claude Code is powerful out of the box, but complex tasks — scaffolding a Python library, initializing a GitHub repo with CI/CD, diagnosing flaky tests, shipping a product idea end-to-end — require careful prompting and domain knowledge that gets lost between sessions. You end up re-explaining the same workflows, catching the same mistakes, and missing the same steps. Skills fix that. Each skill is a directory of markdown files that Claude loads as context when you invoke it. The domain knowledge, workflow steps, and quality rules are encoded once and reused every time.

How a skill is structured

Every skill is a directory containing a SKILL.md entrypoint and optional reference files. SKILL.md is the table of contents for the skill. It contains:
  • Numbered workflow steps, each pointing to a reference file for details
  • Golden rules — hard, mechanical constraints that prevent common mistakes
  • A self-review checklist Claude runs before delivering output
Reference files contain the detailed guidance: templates, code patterns, examples, and domain knowledge. They are loaded progressively — only when Claude needs them — so context stays focused. This architecture has four key properties:
  • Table-of-contents structureSKILL.md stays concise (~50–100 lines). Details live one level deeper in reference files.
  • Progressive disclosure — Only SKILL.md loads on invocation. Reference files load on demand, keeping the context window efficient.
  • Golden rules — Each skill encodes 3–8 hard mechanical rules that prevent the most common failure modes for that domain.
  • Feedback loops — Every skill includes at least one do → check → fix cycle, so errors are caught and corrected before output is delivered.

Skill categories

Building & Shipping

Ship product ideas end-to-end, scaffold Python libraries, initialize GitHub repos with CI/CD, and build digital board games.

Research & Design

Multi-dimensional research, adversarial design review, gap finding, spec writing, and competitive analysis.

Visualization

Render interactive charts, diagrams, slide decks, sortable tables, and Gantt timelines live in the browser.

Debugging & Quality

Diagnose flaky tests with structured multi-run experiments and scan codebases for 24 production readiness items.

Creating Skills

Build new skills using harness engineering best practices: table-of-contents architecture, golden rules, and self-review loops.

Magic Fetch

Log every capability gap in real-time and synthesise them into a prioritised integration roadmap.

Next steps

Installation

Clone the repo and link skills to your Claude Code config in under two minutes.

Quick Start

Run your first skill and understand how the slash-command system works.