Install the Agent Developer Kit for declarative agents with the one-shot installer

  • Applies to: ✅ Microsoft Copilot

The one-shot installer sets up the required tools, clones the kit, installs the GitHub Copilot Chat extension, and opens the ess-maker-skills folder in Visual Studio Code.

Note

The one-shot installer supports Windows and macOS. If you're on Linux, or you can't install software locally, use GitHub Codespaces instead.

Before you start

You need:

  • A Windows 10 or Windows 11 machine, or a Mac, where you can install software.
  • A GitHub account with an active GitHub Copilot subscription. If you aren't sure, your administrator can confirm in the GitHub organization settings.
  • Permission to run PowerShell or Terminal with internet access.

Run the installer

Windows

  1. Open PowerShell (Windows key, type powershell, select Windows PowerShell).

  2. Paste this command and select Enter:

    iex (irm https://raw.githubusercontent.com/microsoft/Employee-Self-Service-Agent-Developer-Kit/main/setup/bootstrap.ps1)

macOS

  1. Open Terminal.

  2. Paste this command and select Enter:

    /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/microsoft/Employee-Self-Service-Agent-Developer-Kit/main/setup/bootstrap-mac.sh)"

The installer prints what it is about to do, then runs through the steps. You may see prompts asking you to confirm software installs. Approve each one. This installer installs the main branch for native declarative agents. Don't use it for Custom Engine Agents or classic declarative-agent preview. For those agent families, use GitHub Codespaces with main-ca or manual installation and select that branch.

Choose Maker or Developer mode

Partway through, the installer asks in the terminal which experience you want:

  • Maker is chat-first. Developer surfaces are hidden, and a Quick Actions rail gives you big buttons for common tasks along with a short built-in tutorial.
  • Developer is the standard Visual Studio Code layout with Copilot Chat in the side panel.

You can switch between the two at any time from the Quick Actions panel.

What the installer does

While it runs, the installer:

  • Installs Visual Studio Code, Python 3.12, Git, and GitHub CLI, plus PowerShell 7 on Windows. It uses winget on Windows and Homebrew on macOS. Tools already installed are skipped.
  • Installs the Python packages the kit needs.
  • Clones the Employee Self-Service Agent Developer Kit repo into your user folder.
  • Installs the GitHub Copilot, GitHub Copilot Chat, and Python extensions in VS Code.
  • Opens VS Code on the solutions/ess-maker-skills folder. In Copilot Chat, run /setup to prepare the agent workspace.

The installer is idempotent. If something interrupts it, run the same command again. It picks up where it left off and skips work already done.

Sign in to GitHub Copilot

When VS Code opens, the bottom-right corner shows a sign-in prompt for GitHub Copilot.

  1. Select Sign in to GitHub.
  2. Your browser opens to GitHub.
  3. Authorize VS Code to use your GitHub account.
  4. The browser returns you to VS Code. The account icon in the lower left now shows your GitHub username.

If VS Code reports that your account doesn't have Copilot access, confirm your subscription at github.com/settings/copilot.

You should see

When the installer finishes:

  • VS Code is open with ESS-MAKER-SKILLS at the top of the Explorer pane.
  • The Explorer shows .github, .vscode, scripts, and src as the top folders.
  • The Copilot Chat pane is available from the activity bar on the left.

If the Explorer shows EMPLOYEE-SELF-SERVICE-AGENT-DEVELOPER-KIT at the top with extra folders (samples, solutions, tests), the installer opened the wrong folder. Close the workspace and reopen solutions/ess-maker-skills.

The kit asks the model to detect tools, follow multi-step instructions, and write valid YAML. Not all models do these actions equally well.

Open the Copilot Chat pane (the Chat icon on the activity bar). On the top of the chat pane, there's a model picker. Pick a current frontier model from one of these families:

Model family Why
Claude Sonnet (latest available) Recommended. Reliable tool detection and multi-step instruction-following.
Claude Opus (latest available) Recommended. Same strengths as Sonnet; higher quality, slightly slower.
OpenAI Codex or GPT-5 class (latest available) Recommended. Works well for the kit's setup flows.

Note

The models in the picker change over time as new versions ship. Pick the most recent version available in one of the previously listed families rather than an exact version number.

If you don't see any of these families in the picker, they may not be enabled by your organization. Choose the closest equivalent. If the kit later reports that it can't find its tools, the most common fix is to switch to a recommended model.

Next step

For native declarative agents, run /setup as described in Set up your workspace. For Custom Engine Agents or classic declarative-agent preview, continue with Run your first command.

If the installer didn't finish, see Troubleshoot getting started.

  • Last updated on