Skip to content

overview.md

Latest commit

 

History

History
74 lines (46 loc) · 4.31 KB

overview.md

File metadata and controls

74 lines (46 loc) · 4.31 KB
title Get started with {% data variables.product.prodname_copilot_short %} agents on {% data variables.product.github %}
shortTitle Get started
intro Try {% data variables.copilot.copilot_cloud_agent %} end-to-end in about ten minutes.
product {% data reusables.gated-features.copilot-cloud-agent %}<br><a href="https://github.com/features/copilot/plans?ref_product=copilot&ref_type=purchase&ref_style=button" target="_blank" class="btn btn-primary mt-3 mr-3 no-underline"><span>Sign up for {% data variables.product.prodname_copilot_short %}</span> {% octicon "link-external" height:16 %}</a>
allowTitleToDifferFromFilename true
versions
feature
copilot
contentType how-tos

Prerequisite

You need a repository where {% data variables.copilot.copilot_cloud_agent %} is enabled. For {% data variables.copilot.copilot_business_short %} and {% data variables.copilot.copilot_enterprise_short %} subscribers, an administrator must enable the agent before you can use it. See AUTOTITLE.

Step 1: Assign an issue to {% data variables.product.prodname_copilot_short %}

Pick a repository where {% data variables.copilot.copilot_cloud_agent %} is enabled and find an open issue—or create a small one, such as "Add a CONTRIBUTING.md file."

  1. In the right sidebar, click Assignees, then select Copilot.

  2. Optionally, add instructions in the Optional prompt field.

    For example: Keep the file short and include a code of conduct section.

  3. Click Assign.

{% data variables.product.prodname_copilot_short %} starts a session and begins working on a pull request.

Step 2: Start a research task at the same time

While {% data variables.product.prodname_copilot_short %} works on the issue, start a second, non-coding task to see two sessions running in parallel.

  1. In the same repository, click the {% octicon "agent" aria-hidden="true" aria-label="agent" %} Agents tab.

  2. Type a research prompt.

    For example: Investigate which dependencies in this repo are outdated and summarize what upgrading would involve.

  3. Click {% octicon "paper-airplane" aria-label="Start task" %} or press Enter.

A second session appears in the panel alongside the first.

Step 3: Watch both sessions

Both sessions update in real time. Click either session to view the agent's reasoning, the files it reads, and the changes it makes.

While a session is running, type in the prompt box below the log to steer the agent.

For example: Focus on security-related dependencies first.

For more details on monitoring, steering, and stopping sessions, see AUTOTITLE.

Step 4: Request a {% data variables.copilot.copilot_code-review_short %} on the pull request

When the coding task finishes, {% data variables.product.prodname_copilot_short %} opens a pull request and adds you as a reviewer.

  1. Open the pull request from the notification or from the session log.
  2. In the right sidebar, click Reviewers, then select {% data variables.product.prodname_copilot_short %}.
  3. Wait for {% data variables.copilot.copilot_code-review_short %} to leave comments—usually under 30 seconds.

Read through the review comments. {% data variables.product.prodname_copilot_short %} may suggest changes you can apply directly with a click.

For the full set of code-review options, see AUTOTITLE.

Step 5: Review the output and iterate

Now review the code changes yourself, just as you would for any contributor's pull request.

  • Request changes from {% data variables.product.prodname_copilot_short %} — mention @copilot in a comment describing what to fix. {% data variables.product.prodname_copilot_short %} pushes new commits to the same branch.
  • Make changes yourself — check out the branch and push your own commits.
  • Approve and merge — when you're satisfied, merge the pull request.

For more on reviewing {% data variables.product.prodname_copilot_short %}'s work, see AUTOTITLE.

Next steps

  • AUTOTITLE — Write effective prompts and get the most out of agents.