Introducing the Agents tab in your repository 🗳️

[ebndev]

We've redesigned where you manage Copilot coding agent tasks. We're bringing a new look and feel to your mission control style view directly into your repository with a new Agents tab.

Instead of navigating to a separate page, your agent sessions now live alongside your code, pull requests, and issues. View all sessions for the repository in one place, create new ones, and switch between tasks without leaving your codebase. One-click links make it easy to jump straight to the associated pull request. You can also archive sessions to keep things organized and page through your session history.

Redesigned session logs

Session logs are now easier to read and follow. Similar tool calls are grouped together, reducing noise and making the flow more clear. Tool outputs render inline previews, and file changes appear with the familiar diff views that you can expand with one click. Every tool call has a distinctive icon, and bash commands are now visible for full transparency.

Resume with the Copilot CLI

If you prefer working in your terminal, you can now continue your Copilot session directly in the Copilot CLI. Click Continue in Copilot CLI, copy the command, and pick up right where you left off.

To get started

1. Navigate to a repository.
2. Click on the new Agents tab.
3. Create a task.
4. To open your session in the CLI, click Continue in Copilot CLI.
5. Copy the command and paste into your terminal to get started with the Copilot CLI.

Note: To view the Agents tab, ensure that you have Copilot coding agent enabled for your repository.

🌟Leave a comment!

Join the discussion and leave feedback in the comments below!

[3kh0]

How do I disable this tab?

[elliebennett]

Hi @3kh0 - The Agents tab appears when Copilot coding agent is enabled for your repository. If it is not enabled, it will not appear.

[3kh0]

I see. I am not sure how it got enabled, but I would think that having a checkbox (under the main settings page -> features, like Wiki and Issues) would be a bit easier to access.

[mkushakov]

Even if Copilot is enabled for repository, not all users are having Copilot licenses. It should be a toggle in repository settings similar to Issues, Wikis and Discussions to turn it on or off
https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/enabling-features-for-your-repository/disabling-issues

[elliebennett]

@mkushakov Thanks for the feedback on how to make the settings more apparent for you and users like you. The Agents tab also has conditional logic based on if the individual user has a Copilot license.

[Plasma]

Thanks for the update, I was hoping to start seeing something about defining recurring agent tasks.

For example, to replace Dependabot, "Keep these packages up to date", or "Do this code scan looking for X issues and raise a Security alert or open a PR", run daily/weekly, etc.

[elliebennett]

Thanks @Plasma. We have something in the work on this. If you're interested in providing early feedback, let me know if you would be open to a feedback session to provide your input.

[katorly]

For anyone who finds this annoying, disable it account-wide in https://github.com/settings/copilot/coding_agent, or organization-wide in https://github.com/organizations/<your_org_name>/settings/copilot/coding_agent.

[SteffenBlake]

Im surprised theres only 3 comments on this

I wanna thank yall for the fact this rollout also seems to correspond with fixing a very annoying bug that has been frustrating me for the past few weeks, agents not getting their agent files loaded in subsequent calls

That bug seems to be fixed now and the improvement to how much "smarter" agents feel as a result is palpable, thanks!

I quite like the new UI, though yall did drop the feature of being able to click a specific LOC during the changes to auto copy it into the chat as FileName#nn:nn to quickly point out a line of code while steering the agent

Not a mission critical feature, but definitely one I noticed was missing right away as I use it quite a lot, heres hoping it makes its way back in though, as I loved that one!

[elliebennett]

Thanks for the feedback @SteffenBlake . We're iterating on the best way to enable you to do this. I'll circle back with you when we have an update! Please continue to share feedback, it helps us improve the experience.

[Inscure]

A feature to disable code collapsing so that all code is visible at once would be useful. Currently, I have to keep clicking "Show all X lines", which is a significant inconvenience.

[elliebennett]

Noted! Thank you @Inscure

[angelo-loria]

Putting this new tab directly in the middle of the most heavily used tabs is diabolical

[garyburgmann]

Right next to Actions is the work of the devil himself (we don't have discussions enabled).

All my tiny brain sees is A***s and I must have accidentally clicked it at least 972 times already.

[dbeattie71]

Configure Environment uses the path repo/.github/workflow/copilot-setup-steps.yml is that correct? Docs say to use repo/github/copilot-setup-steps.yml

[Inscure]

Thanks for building great software for developers. I’d like to share some feedback regarding the recent changes.

I was disappointed to see that the previously available view of modified files — similar to the Changes tab on pull request pages — was removed from the agent session page. With the new agent session page, viewing the summary of code changes prepared by the agent now requires opening the pull request in a separate web browser tab, which feels less convenient. While these changes may have a rationale internally, they were unexpected for me and have had a negative impact on my workflow.

I also noticed that the agent session window appears to use only about 70% of the available horizontal space, leaving unused space on both the left and right sides. This space could potentially be used to widen the session view or to reintroduce a file change preview, as was available before.

One additional usability issue I’d like to mention concerns text input on tablets. Pressing Enter currently submits the prompt to the agent instead of inserting a new line. This differs from the behavior on platforms like the ChatGPT website, where Enter creates a new line and submission is handled separately. As a result, I’ve accidentally submitted incomplete prompts multiple times and had to consume additional premium requests unnecessarily. Adjusting this behavior would help prevent accidental submissions and make writing longer or more structured prompts on tablets much more comfortable.