Introduction
You can use Copilot cloud agent (formerly Copilot coding agent) to:
- Perform deep research on a repository by asking Copilot questions.
- Plan and refine an approach before Copilot makes changes.
- Iterate on code changes on a branch, review a diff, and get Copilot to only create a pull request when you're ready.
Copilot Codierungsassistent won't automatically open a pull request as part of the session. If you do want Copilot to open a pull request for your task right away, you can explicitly ask for one in your prompt. For example, by stating "Create a pull request to ...".
Hinweis
Deep research, planning, and iterating on code changes before creating a pull request are only available with Copilot Codierungsassistent on GitHub.com. Programmier-Agent integrations (such as Azure Boards, JIRA, Linear, Slack, or Teams) only support creating a pull request directly.
Performing deep research
You can ask Copilot Codierungsassistent questions about a repository to understand how it works, identify where to make a change, or confirm assumptions before planning and implementation. Copilot starts a deep research session to examine the repository and return a grounded answer.
-
Öffnen des Agenten-Panels oder der Registerkarte:
- Öffnen Sie die Registerkarte Agents in einem Repository.
-
**Navigieren zur Seite „Agents“**: Wechsle zu [github.com/copilot/agents](https://github.com/copilot/agents?ref_product=copilot&ref_type=engagement&ref_style=text). Alternativ kannst du zu dieser Seite navigieren, indem du das Panel „Agents“ öffnest und dann auf **View all** klickst. -
**Öffnen des Panels „Agents“**: Klicke auf der Navigationsleiste oben rechts auf GitHub auf <svg version="1.1" width="16" height="16" viewBox="0 0 16 16" class="octicon octicon-agent" aria-label="The Agents icon" role="img"><path d="M14.5 8.9v-.052A2.956 2.956 0 0 0 11.542 5.9a.815.815 0 0 1-.751-.501l-.145-.348A3.496 3.496 0 0 0 7.421 2.9h-.206a3.754 3.754 0 0 0-3.736 4.118l.011.121a.822.822 0 0 1-.619.879A1.81 1.81 0 0 0 1.5 9.773v.14c0 1.097.89 1.987 1.987 1.987H4.5a.75.75 0 0 1 0 1.5H3.487A3.487 3.487 0 0 1 0 9.913v-.14C0 8.449.785 7.274 1.963 6.75A5.253 5.253 0 0 1 7.215 1.4h.206a4.992 4.992 0 0 1 4.586 3.024A4.455 4.455 0 0 1 16 8.848V8.9a.75.75 0 0 1-1.5 0Z"></path><path d="m8.38 7.67 2.25 2.25a.749.749 0 0 1 0 1.061L8.38 13.23a.749.749 0 1 1-1.06-1.06l1.719-1.72L7.32 8.731A.75.75 0 0 1 8.38 7.67ZM15 13.45h-3a.75.75 0 0 1 0-1.5h3a.75.75 0 0 1 0 1.5Z"></path></svg>.
-
Ask a question about the repository.
For example:
Investigate performance issues in this app and suggest improvements.Tipp
Alternatively, open Copilot-Chat and attach the relevant repository as context, then ask Copilot-Chat to do research. You will be prompted to approve a deep research session.
-
Optionally, send additional prompts while Copilot is working to steer the session.
-
Review the response and ask follow-up questions.
Creating a plan
You can ask Copilot Codierungsassistent to propose a plan before making changes, simply by prompting the agent.
-
Ask the agent to create a plan.
For example:
Create a plan to implement the most impactful performance improvements for this app. -
Review the plan and iterate with Copilot until it matches what you want.
Iterating on code changes
After research and planning, you can make changes on a branch, review the diff, and then decide whether to iterate more or create a pull request.
-
Ask the agent to implement changes by describing what you want Copilot to do in your prompt.
For example, you might reference the plan you designed earlier:
Implement the plan we agreed upon. -
Copilot starts a session to make the required changes.
-
Once the session completes, click Diff to review the changes.
-
Optionally, open the branch (
copilot/BRANCH-NAME) that Copilot created to inspect the changes in context. -
If you want refinements, you can continue the conversation and ask the agent to adjust the work. For example:
Rename this to match our existing conventions. -
Continue iterating until you’re satisfied with the branch.
-
Once you're happy with the results, and when the session is complete, click Create pull request.