GitHub Copilot can help while you work in an editor, and its cloud agent can take an assigned task through to a proposed change. Those are different ways of working. In the editor, you are working in your local project. In the cloud, the agent needs a task, a repository, and an environment in which it can complete the work.
Start with the editor route if you want to understand each step of your first contribution. Use cloud delegation once you can describe the outcome and evaluate the returned work.
Set up Copilot in VS Code
First clone the repository, open it in VS Code, and create a contribution branch. Confirm Git works independently of the assistant. Then follow Microsoft's Copilot setup guide: use the Copilot status control, enable AI features, and complete the sign-in flow for the account that holds your entitlement.
Use your organization-provided account for managed work. The setup guide also documents a Continue with GHE.com route for applicable enterprise accounts. That is not interchangeable with an arbitrary GitHub Enterprise Server URL. Repository hosting, account identity, and Copilot entitlement need to match the supported configuration.
Once the assistant opens, ask it to identify the repository and explain the README's installation instructions. This is a more useful first check than asking for a generic explanation of GitHub. You want evidence that it is using your project context.
Use a small, complete task
The example below asks for a documentation change and defines the completion evidence:
Help improve README.md for a first-time user downloading this project.
Read applicable contribution instructions and the existing installation
section. Clarify which documented route provides an installable package
and which provides source code.
Do not invent release assets or platform support. If the repository does
not identify a distribution, report that gap. Keep the change to README.md.
Show the proposed wording, apply it after review, and report checks actually
performed. Leave Git commits and publication to me for this first task.Use the assistant's editing-capable mode when you are ready to apply the change. Inspect its proposed tool operations and the resulting diff. If it needs a command solely to format or preview Markdown, that is different from installing an application dependency tree. Keep the work proportional to the task.
After the edit, inspect Source Control. The intended result is a small README change, with no accidental modifications to configuration or generated files. Preview the paragraph and verify the linked installation route. Commit and publish the branch, then open the pull request in GitHub or through the pull-request extension.
If Copilot supplies a test summary, check what it actually means. “No code behavior changed” is a statement about scope; “all tests passed” requires executed tests. A clean-looking paragraph should not acquire a fictional validation history on its way to the reviewer.
Delegate an issue to the cloud agent
When available and enabled for your account and repository, GitHub's cloud-agent workflow can start from assigning an issue to Copilot. GitHub's current cloud-agent instructions describe issue assignment as a public-preview feature. Verify availability in your environment rather than promising it for every GitHub installation.
- Open or create a real issue describing the documentation problem.
- Include the target file, the reader's difficulty, the desired outcome, and acceptance criteria.
- Choose Copilot in the issue's assignee control when it is available.
- Confirm the destination repository and starting branch in the assignment options.
- Follow the agent's resulting pull request and review its changes and checks.
For this example, use an issue title such as Clarify the intended download in the README. The body should require the existing installation route to remain accurate, distinguish source archives from installable distributions where relevant, and restrict the change to documentation.
After assignment, direct follow-up information to the generated pull request. GitHub's documentation notes that later issue comments are not automatically included in the task context. Keeping the working conversation beside the proposed change makes the correction visible to both the agent and reviewers.
When the expected option is missing
A missing assistant or assignee can reflect entitlement, organization policy, repository configuration, product availability, or the selected account. It does not mean you should grant yourself broader repository permissions or move work to another host.
For editor problems, first prove that the correct account can use Copilot and that the intended folder is open. For cloud problems, check whether the feature is enabled for that repository and whether the environment can perform the required setup. For a push failure, return to Git authentication and branch permissions.
Use Copilot to reduce the effort of preparing a contribution. Preserve a clear review boundary: the change, its reason, and the verification should remain understandable to someone who never saw the assistant conversation.