Creating Workflows

GitHub Agentic Workflows provide a custom agent that turns GitHub Copilot into a powerful workflow authoring tool. This guide explains how to use this agent to author agentic workflows in natural language.

Quick Start

Initialize your repository to set up GitHub Copilot instructions and custom agents for authoring agentic workflows:

gh aw init

This creates:

• .github/instructions/github-agentic-workflows.instructions.md - Custom Copilot instructions for better workflow authoring
• .github/agents/create-agentic-workflow.md - Custom agent for the /agent command in Copilot Chat
• .github/agents/create-shared-agentic-workflow.md - Custom agent for creating reusable workflow components

VS Code integration

Once initialized, GitHub Copilot will automatically use these instructions when you edit workflow files in VS Code. See VS Code Integration for more details.

What is the create-agentic-workflow custom agent?

.github/agents/create-agentic-workflow.md is a custom agent file that contains the structure and instructions GitHub Copilot will use to generate a workflow markdown file that the gh aw CLI understands.

Use the custom agent when you want to:

• Draft a new agentic workflow using natural language
• Iterate on workflow steps with AI assistance inside your editor

The custom agent contains specialized instructions and toolset to enable efficient workflow authoring.

To get this file in your repository, run the init command:

gh aw init

How to use the create-agentic-workflow custom agent

GitHub Copilot Chat in Visual Studio Code

In Visual Studio Code with GitHub Copilot Chat, use the /agent command and select create-agentic-workflow from the list:

/agent

Then select create-agentic-workflow from the available custom agents. This will start the agentic workflow authoring process with interactive guidance.

GitHub Copilot CLI

Assuming you have the GitHub Copilot CLI installed, you can reference the custom agent file using the --agent flag:

copilot --agent .github/agents/create-agentic-workflow.md

Other Agents and Chats

The custom agent file can be loaded into other AI chat interfaces that support custom instructions. The agent is designed to be compatible with various AI tools, although some features might require configuration and you’ll need to allow running the compiler.

Dictating Agentic Workflows

When creating agentic workflows using speech-to-text (dictation), you may encounter terminology mismatches and formatting issues common to voice recognition systems. To help correct these issues, use the dictation instructions prompt or Copy full instructions .

This prompt helps rephrase text captured through speech-to-text recognition by:

• Correcting project-specific terminology (e.g., “ghaw” → “gh-aw”, “work flow” → “workflow”)
• Transforming casual dictated sentences into clear, imperative task descriptions
• Removing filler words and improving technical tone
• Adding necessary context that might be implicit in spoken requests

Load the dictation prompt into your AI assistant before or after dictating your workflow instructions to improve accuracy and clarity.

Designing Tasks with ChatGPT

When designing agentic workflow tasks, you can use the agentic-chat instructions with ChatGPT or other conversational AI applications to create clear, actionable task descriptions. Use the button below to copy the full instructions: Copy agentic-chat instructions .

The agentic-chat assistant helps you:

• Break down complex problems into clear, actionable phases
• Structure task descriptions for AI coding agents
• Write specifications without implementation details
• Follow best practices for agentic workflow task descriptions

To use the agentic-chat instructions:

1. Copy the full instructions using the button above
2. Paste them into your ChatGPT or other AI chat interface
3. Describe your workflow goal or problem
4. The assistant will ask clarifying questions and generate a structured task description
5. Copy the generated task description (wrapped in 5 backticks) and use it in your workflow

The assistant uses a neutral, technical tone and focuses on what needs to be done rather than how to implement it, making it ideal for creating specifications that AI coding agents can execute.

After compiling

Commit the generated files (.lock.yml) if they are part of the project’s tracked artifacts. The project uses compiled workflows in version control.