How to Add a Specific Garden Skill to an Agent: Complete Setup Guide

To add a specific Garden Skill to an agent, copy the skill folder from skills/<skill-name>/ into your agent's ./.agents/skills/<skill-name>/ directory and optionally register it in the agent's YAML configuration file.

Garden Skills are self-contained skill packages that extend the capabilities of AI coding agents in the ConardLi/garden-skills repository. Each skill defines a specific workflow through its SKILL.md contract and can be imported on-the-fly by compatible agents. Learning how to add a specific Garden Skill to an agent enables you to customize your AI assistant's behavior for specialized tasks without modifying core agent code.

Step 1: Select the Skill from the Repository

Browse the available skills in the skills/ directory of the ConardLi/garden-skills repository. Each skill is organized in its own subfolder, such as skills/web-design-engineer/ or similar specialized packages. The repository's README provides a complete skill index showing all available options and their capabilities.

Every skill contains two critical files that define its behavior:

  • SKILL.md: The formal contract describing the skill's workflow, phases, checkpoints, and required inputs.
  • manifest.json: Optional metadata including version, author, and description for tooling integration.

Step 2: Copy the Skill to the Agent's Skills Directory

Agents look for skills in one of two conventional locations within your project:

  • ./.agents/skills/<skill-name>/
  • ./.claude/skills/<skill-name>/

The Manual copy method is the most straightforward way to add a specific Garden Skill to an agent. Simply copy the entire skill folder from the repository into one of these directories.

Execute the following commands in your terminal:


# Define the skill you want to add

SKILL=web-design-engineer

# Create the target directory and copy the skill files

mkdir -p .agents/skills/${SKILL}
cp -r skills/${SKILL}/* .agents/skills/${SKILL}/

Once copied, the agent will automatically load the skill the next time it runs.

Step 3: Register the Skill in the Agent Configuration (Optional)

While the agent will detect skills placed in the skills directory, explicitly registering the skill in your agent's YAML manifest provides better control and allows you to set skill-specific defaults.

Add the skill name to the skills: list in your agent configuration file, typically located at .agents/agents/openai.yaml:


# .agents/agents/openai.yaml

skills:
  - web-design-engineer   # Newly added skill

The skill folder contains its own agents/openai.yaml with sensible defaults. Copying this file into your project-wide agents directory merges these defaults with your custom configuration, ensuring the skill behaves according to its intended parameters.

Understanding the Garden Skills File Structure

Recognizing these core files helps you troubleshoot and customize skill integration:

  • skills/<skill-name>/SKILL.md: Defines the formal contract that tells the agent how to execute the skill workflow, including mode detection and checkpoint sequences.
  • skills/<skill-name>/manifest.json: Stores metadata such as version, author, and description, used by tooling that lists available skills.
  • README.md: Located in the repository root, this file lists all available skills and documents the three installation methods, including the manual copy approach.
  • .agents/agents/openai.yaml: The agent-wide configuration file where you explicitly enable skills and override default behaviors.

Summary

  • Garden Skills are self-contained packages stored in skills/<skill-name>/ that extend agent capabilities through formal contracts defined in SKILL.md.
  • To add a specific Garden Skill to an agent, manually copy the skill folder to ./.agents/skills/<skill-name>/ or ./.claude/skills/<skill-name>/.
  • The Manual copy method requires no additional tooling—simply copy the directory contents and restart the agent.
  • Optional explicit registration in .agents/agents/openai.yaml provides control over skill defaults and makes the skill explicitly available in the agent's skill registry.
  • Agents automatically load available skills on startup, requiring no code changes to use the new functionality.

Frequently Asked Questions

Where does the agent look for Garden Skills?

According to the ConardLi/garden-skills source code, agents search for skills in two conventional locations: ./.agents/skills/<skill-name>/ and ./.claude/skills/<skill-name>/. Placing the skill folder in either location makes it available for automatic loading the next time the agent initializes.

Do I need to modify agent code to use a new Garden Skill?

No. Once you copy the skill files into the appropriate skills directory, the agent automatically detects and loads the skill the next time it runs. No code changes, recompilation, or package installation steps are required, as the discovery mechanism scans the conventional paths at startup.

What is the difference between SKILL.md and manifest.json?

SKILL.md contains the operational contract that defines how the agent executes the skill, including workflow phases, checkpoints, and required inputs. manifest.json contains metadata such as version number, author information, and description tags, primarily used by listing tools and package managers rather than the runtime agent execution engine.

Can I override default skill settings when adding a Garden Skill?

Yes. While the skill folder includes default configurations in its internal agents/openai.yaml, you can override these by registering the skill in your project-wide .agents/agents/openai.yaml file. This explicit registration allows you to customize skill-specific parameters while maintaining the core workflow defined in SKILL.md.

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:

Share the following with your agent to get started:
curl -s "https://instagit.com/install.md"

Works with
Claude Codex Cursor VS Code OpenClaw Any MCP Client

Maintain an open-source project? Get it listed too →