How to Activate a Skill in Claude.ai: The Complete Setup Guide
You can activate a skill in Claude.ai by installing the skill files into the agent’s skills/ directory via the Claude Code plugin marketplace, the npx skills CLI, or manual copy, after which Claude automatically loads the SKILL.md system prompt.
Activating a skill transforms Claude.ai from a general-purpose assistant into a domain-specific expert. In the ConardLi/garden-skills repository, each skill is a structured bundle of prompts and metadata that guide Claude’s behavior for specialized tasks like web design or article authoring.
What Is a Claude.ai Skill?
A Skill is a directory containing a SKILL.md file that defines the system prompt, a manifest.json for version metadata, and documentation. According to the garden-skills source code, the skill architecture relies on three core components:
SKILL.md– Located atskills/<skill-name>/SKILL.md, this file contains the markdown front-matter header and the complete system prompt that reshapes Claude’s behavior.manifest.json– Found in each skill directory, this file declares the version and lists files to copy during installation.README.md– Human-readable documentation guiding users through activation steps and usage patterns.
When these files reside in the agent’s skills/ directory, Claude automatically reads the SKILL.md on startup and applies the system prompt to subsequent generations.
Prerequisites for Skill Activation
Before activating a skill, ensure you have access to Claude Code or a Claude.ai environment that supports the skills protocol. You will need one of the following:
- Claude Code console – The terminal-based agent interface that supports the
/plugincommand. - Node.js and npm – Required only if using the npx CLI installation method.
- Git – Necessary for manual installation via submodules or direct cloning.
Three Methods to Activate a Skill in Claude.ai
The garden-skills repository supports three distinct installation paths, each suited to different workflows.
Method 1: Plugin Marketplace (Claude Code)
The fastest way to activate a skill in Claude.ai is through the built-in plugin marketplace. This method fetches the skill directly from the garden-skills repository and installs it into your agent’s environment.
# Inside the Claude Code console
/plugin install web-design-engineer@garden-skills
Claude immediately loads skills/web-design-engineer/SKILL.md and integrates the design-engineer system prompt. No restart is required; the skill activates as soon as the command completes.
Method 2: npx CLI Installer
For users who prefer terminal-based workflows or need to script installations, the universal skills CLI provides a one-line installer. This tool reads the manifest.json, downloads the necessary files, and places them in the correct skills/ subdirectory.
# One-liner that fetches and installs the skill
npx skills add web-design-engineer
The CLI handles version resolution using the metadata in skills/web-design-engineer/manifest.json, ensuring you receive the correct file set without manual copying.
Method 3: Manual Copy or Git Submodule
Advanced users who require version pinning or customization can install skills manually. This approach involves copying the skill directory directly into the agent’s skills/ folder or tracking it as a Git submodule.
# Option A: Add as a Git submodule
git submodule add https://github.com/ConardLi/garden-skills.git garden-skills
# Option B: Copy only the specific skill
cp -r garden-skills/skills/web-design-engineer ./skills/web-design-engineer
After copying, verify that SKILL.md exists at ./skills/web-design-engineer/SKILL.md. Claude detects the file automatically on the next interaction.
How Skill Activation Works Under the Hood
When you activate a skill using any of the three methods, the same loading mechanism occurs. The Claude.ai agent scans its skills/ directory for subdirectories containing SKILL.md files. Upon detecting a new skill, the agent:
- Parses the YAML front-matter in
SKILL.mdto extract metadata. - Loads the system prompt body defined in the markdown.
- Appends the prompt to the context window for subsequent requests.
This process requires no code changes or configuration files beyond placing the skill files in the correct location. The activation is stateless; removing the skill directory immediately deactivates the behavior.
Verifying Your Skill Is Active
To confirm successful activation, issue a prompt that falls within the skill’s domain. For the web-design-engineer skill installed from the garden-skills repository:
User: Design a responsive landing page for a travel app.
If the skill is active, Claude will respond with HTML/CSS that follows the specific design system defined in skills/web-design-engineer/SKILL.md, rather than generic styling advice. The response structure and terminology will align with the specialized prompt instructions.
Summary
- Skill Structure: Each skill requires
SKILL.md(system prompt),manifest.json(metadata), and documentation in a directory underskills/<skill-name>/. - Activation Methods: Use
/plugin installfor Claude Code,npx skills addfor CLI automation, or manual copy/Git submodule for full control. - Automatic Loading: Claude.ai reads
SKILL.mdautomatically once files are present in the agent’sskills/directory—no restart or additional configuration required. - Source Location: All examples reference the ConardLi/garden-skills repository, specifically paths like
skills/web-design-engineer/SKILL.mdandscripts/release/pack-skill.mjs.
Frequently Asked Questions
What file actually activates the skill in Claude.ai?
The SKILL.md file triggers activation. Located at skills/<skill-name>/SKILL.md, this file contains the system prompt that Claude loads into its context. As long as this file exists in the agent’s skills directory, the skill is active regardless of how it was installed.
Can I use multiple skills simultaneously?
Yes. Claude.ai supports concurrent skill activation. Install multiple skills using any combination of the three methods, and Claude will load each SKILL.md it discovers in the skills/ directory. Be aware that overlapping system prompts may cause conflicting behaviors if skills define contradictory instructions.
Do I need to restart Claude after installing a skill?
No. Claude detects new skill directories automatically on the next interaction. Once the SKILL.md file is present in the correct location, the system prompt loads immediately for subsequent prompts without requiring a process restart.
Where does the npx CLI store skills locally?
The npx skills CLI copies files into the agent’s local skills/ directory according to the paths defined in each skill’s manifest.json. You can verify the installation by checking that skills/<skill-name>/SKILL.md exists in your project root or Claude Code working directory.
Have a question about this repo?
These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:
curl -s "https://instagit.com/install.md" Maintain an open-source project? Get it listed too →