GHub Copilot gives you the option to create your own agents (before called chatmodes). Such a custom agent can pick up specific tasks and take on a role inside the development process e.g. a code reviewer, a tester, an architect, ...
Custom agents are defined in an .agent.md Markdown file and consists of the following elements:
- A name
- A description
- An (optional) model
- A set of tools that the agent can use
- And of course, a prompt
Remark: For the full list of elements, have a look at the documentation here.
Here is an example from the Awesome Copilot Github repo:
Handoffs
With the latest VSCode update an extra element was introduced, handoffs. Handoffs allow to create sequential workflows where after one agent has completed it’s job, it can handover the work to another agent.
This handoff can happen automatically or through handoff buttons that let you choose how to continue.
Let’s us create an example:
- First I create 2 agents
- Notice that in the first agent I added an handoffs section. In this section I can specify one or more handoffs.
- Each handoff has the following values:
- A label: A clear description to identify the next step
- An agent: which agent should go next
- A prompt: A prompt that can be passed to the agent
- A send value: Setting this to true will automatically do the handoff to this agent
- Don’t be afraid to make any mistakes as the VSCode team provided IntelliSense;
- Now we can ask our first agent to go to work:
- Once it completes its task, you get the option to choose how to proceed:
- You can continue in the foreground or delegate work to an agent in the background (more about that in the next post) or in the cloud:
Nice!
More information
Supercharge your AI Workflow with Github Copilot Custom Prompt Files