use-cases

Building Agents With Agent Kernel Skills

This guide shows how to use the Agent Kernel skills pack with a coding agent to create a new agent project. The sample agent directory is waste-sorting-assistant, but the same workflow can be reused with any folder name.

Python version must be 3.12 - 3.13.x.

Before starting, install the agentkernel package first; this installs the ak CLI used to download the skills pack. Follow the Agent Kernel quick-start installation path for your preferred framework.

Install Agent Kernel with:

pip install agentkernel

1. Create Or Enter The Agent Directory

From the directory where you keep agent projects, create the agent folder if it does not already exist, then move into it:

mkdir -p waste-sorting-assistant
cd waste-sorting-assistant

For another agent, replace waste-sorting-assistant with your own folder name:

mkdir -p <your-agent-folder>
cd <your-agent-folder>

2. Define The Agent Spec

Create a SPEC.md file in the agent folder. This file should describe the agent's purpose, tools, memory needs, local run mode, deployment target, and any folder or file structure requirements.

3. Download The Skills Pack

List the available Agent Kernel skills and supported coding assistants:

ak skill list
ak skill assistants

Install the skills for the coding agent you will use. For example:

ak skill install --assistant cursor

Other supported assistants use the same command with a different assistant name:

ak skill install --assistant copilot
ak skill install --assistant claude
ak skill install --assistant windsurf
ak skill install --assistant codex
ak skill install --assistant aider

The coding agent discovers the installed skills from its configured skill directory inside the current agent folder. To have those skills auto-applied, open the agent folder itself as the workspace root, for example use-cases/waste-sorting-assistant. If you keep the larger repository open as the workspace root, nested skill files can still be read manually, but the coding agent may not automatically apply them.

4. Ask The Coding Agent To Build The Agent

Use the installed Agent Kernel skills and the local SPEC.md to scaffold and implement the agent. For the sample waste-sorting-assistant, give the coding agent this prompt:

Using the Agent Kernel skills pack, read SPEC.md and build the Agent Kernel project in this folder.

Follow the requirements in SPEC.md. Create the project structure, dependencies, configuration, agent implementation, tools, memory/session persistence, local run path, tests, and deployment files described there.

5. Continue Development

After the first scaffold is generated, use follow-up prompts such as:

Add automated tests for the disposal lookup tool and memory behavior.

Add a REST API entry point for this agent.

Add deployment configuration for the target cloud provider.

The coding agent should use the relevant Agent Kernel skills, such as ak-init, ak-build, ak-add-capabilities, ak-cloud-deploy, and ak-test, to generate the project files and keep the implementation aligned with Agent Kernel conventions.