Have Copilot coding agent write and push code
Uh oh! Task: Uh oh! There was an error while loading. Please reload this page. Uh oh! There was an error while loading. Please reload this page. wojtekmaj Aug 21, 2025 Select Topic Area Product Feedback Copilot Feature Area Copilot Coding Agent Body I’ve been testing GitHub Copilot Agent recently, and I want to start by saying how impressed I am with the feature. In most cases it does a remarkably good job at understanding context and producing meaningful code contributions. The experience of assigning an issue and immediately seeing the agent respond—even with something as simple as a confirmation emoji—feels smooth and welcoming. Overall, the potential here is huge. That said, I’ve run into some areas where the experience didn’t fully “click.” I’m sharing them below in a structured way. If Copilot instructions aren’t already set up, the agent tends to wander and produce a weak or off-target first PR, or just works slower than it's able to. This often feels like wasted effort. It would be better if the workflow started with an explicit onboarding step that explains: why instructions are important, how to configure them, and what kind of instructions produce the best results. Even something like a short checklist (“Have you configured Copilot instructions in your repo?”) before assigning the agent would make the first PR much more useful. It’s not clear when or how the agent picks up new instructions. For example: If instructions are updated on the default branch, will the agent apply them on a feature branch without manual merging? Is it running with the latest commit, or an older snapshot? An indicator on the agent’s activity (e.g. “Running with instructions from commit abc123”) would make this transparent. Otherwise, it’s easy to lose trust if behavior doesn’t match the latest documented guidance. Similar indicator could be made for whether or not Copilot Setup Steps were picked up, and if so - where from. The agent currently struggles with: PR references: even when linking to PRs in the same org, the agent was not able to access them. Attachments: it couldn’t read attached files when referenced in the issue. Eventually, I had to paste content directly, which worked, but isn’t practical for larger inputs. It would be a big usability improvement if Copilot could handle attachments and org-internal PR links. And if access is blocked for security reasons, a clear error message (“Copilot cannot read attachments in this repository”) would be better than proceeding with flawed assumptions. It wasn’t obvious when the agent would engage during PR reviews. I discovered that explicitly mentioning @copilot in a comment works, but: PR review seems to be ignored, There’s no clear UI signal about what triggers Copilot, A small clarification in the docs (or even in the UI) would help a lot. Ideally, this could be configurable—e.g. a toggle to say “Copilot should respond to review comments by default” vs. “only when explicitly mentioned.” The emoji reaction when an issue is assigned to Copilot is a fantastic little touch. It gives confidence that the agent is working. When the agent gets “stuck” (e.g. can’t read input or doesn’t know what to do), it should fail fast and ask for clarification rather than continue in the wrong direction. Closing Note Overall, I’m very impressed with Copilot Agent. When it works, it feels like a powerful teammate. The feedback above is meant to highlight where the developer experience can be smoother and more predictable. Thanks for the opportunity to try it out and share feedback. (This feedback summary was generated with the help of AI.) Replies: 4 comments · 2 replies Uh oh! There was an error while loading. Please reload this page. github-actions[bot] Bot Aug 21, 2025 💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩ Your input will be carefully reviewed and cataloged by members of our product teams. Due to the high volume of submissions, we may not always be able to provide individual responses. Rest assured, your feedback will help chart our course for product improvements. Due to the high volume of submissions, we may not always be able to provide individual responses. Rest assured, your feedback will help chart our course for product improvements.; tools: GitHub Copilot. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include setting up Copilot instructions in the repository to avoid wandering behavior. Constraints include the agent's inability to access attachments or internal PR links, necessitating manual input. Integration uncertainty exists regarding when the agent picks up new instructions from different branches. Proposed advice: Add a pre-assignment checklist for instruction configuration and implement clear error messages when access is blocked..
Tools
Published on Agent Work Atlas:
Overview
Uh oh!
Workflow
Uh oh! There was an error while loading. Please reload this page. Uh oh! There was an error while loading. Please reload this page. wojtekmaj Aug 21, 2025 Select Topic Area Product Feedback Copilot Feature Area Copilot Coding Agent Body I’ve been testing GitHub Copilot Agent recently, and I want to start by saying how impressed I am with the feature. In most cases it does a remarkably good job at understanding context and producing meaningful code contributions. The experience of assigning an issue and immediately seeing the agent respond—even with something as simple as a confirmation emoji—feels smooth and welcoming. Overall, the potential here is huge. That said, I’ve run into some areas where the experience didn’t fully “click.” I’m sharing them below in a structured way. If Copilot instructions aren’t already set up, the agent tends to wander and produce a weak or off-target first PR, or just works slower than it's able to. This often feels like wasted effort. It would be better if the workflow started with an explicit onboarding step that explains: why instructions are important, how to configure them, and what kind of instructions produce the best results. Even something like a short checklist (“Have you configured Copilot instructions in your repo?”) before assigning the agent would make the first PR much more useful. It’s not clear when or how the agent picks up new instructions. For example: If instructions are updated on the default branch, will the agent apply them on a feature branch without manual merging? Is it running with the latest commit, or an older snapshot? An indicator on the agent’s activity (e.g. “Running with instructions from commit abc123”) would make this transparent. Otherwise, it’s easy to lose trust if behavior doesn’t match the latest documented guidance. Similar indicator could be made for whether or not Copilot Setup Steps were picked up, and if so - where from. The agent currently struggles with: PR references: even when linking to PRs in the same org, the agent was not able to access them. Attachments: it couldn’t read attached files when referenced in the issue. Eventually, I had to paste content directly, which worked, but isn’t practical for larger inputs. It would be a big usability improvement if Copilot could handle attachments and org-internal PR links. And if access is blocked for security reasons, a clear error message (“Copilot cannot read attachments in this repository”) would be better than proceeding with flawed assumptions. It wasn’t obvious when the agent would engage during PR reviews. I discovered that explicitly mentioning @copilot in a comment works, but: PR review seems to be ignored, There’s no clear UI signal about what triggers Copilot, A small clarification in the docs (or even in the UI) would help a lot. Ideally, this could be configurable—e.g. a toggle to say “Copilot should respond to review comments by default” vs. “only when explicitly mentioned.” The emoji reaction when an issue is assigned to Copilot is a fantastic little touch. It gives confidence that the agent is working. When the agent gets “stuck” (e.g. can’t read input or doesn’t know what to do), it should fail fast and ask for clarification rather than continue in the wrong direction. Closing Note Overall, I’m very impressed with Copilot Agent. When it works, it feels like a powerful teammate. The feedback above is meant to highlight where the developer experience can be smoother and more predictable. Thanks for the opportunity to try it out and share feedback. (This feedback summary was generated with the help of AI.) Replies: 4 comments · 2 replies Uh oh! There was an error while loading. Please reload this page. github-actions[bot] Bot Aug 21, 2025 💬 Your Product Feedback Has Been Submitted 🎉 Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users. Here's what you can expect moving forward ⏩ Your input will be carefully reviewed and cataloged by members of our product teams. Due to the high volume of submissions, we may not always be able to provide individual responses. Rest assured, your feedback will help chart our course for product improvements. Due to the high volume of submissions, we may not always be able to provide individual responses. Rest assured, your feedback will help chart our course for product improvements.
Reported result
The author reports that without configured instructions, the agent produces weak or off-target first PRs and works slower than capable. The agent fails to read attachments or access org-internal PR links, requiring manual content pasting. PR review comments are often ignored unless @copilot is explicitly mentioned, leading to unpredictable engagement. No measured success metrics are provided, only qualitative observations of friction and wasted effort.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include setting up Copilot instructions in the repository to avoid wandering behavior. Constraints include the agent's inability to access attachments or internal PR links, necessitating manual input. Integration uncertainty exists regarding when the agent picks up new instructions from different branches. Proposed advice: Add a pre-assignment checklist for instruction configuration and implement clear error messages when access is blocked.
Try it
Act as a developer using GitHub Copilot Coding Agent. I have a repository [replaceable_repo_url] with an issue [replaceable_issue_id] that references an attachment [replaceable_attachment_content] and an internal PR [replaceable_pr_link]. First, verify that Copilot instructions are configured in the repo; if not, pause and ask me to provide them. Next, manually paste the attachment content and PR details into the issue comment to ensure the agent can access them. Assign the issue to the Copilot Agent. Monitor the agent's activity log to confirm it is running with the latest instructions. If the agent produces a PR, review it. If you need the agent to respond to a review comment, explicitly mention @copilot in the comment. Do not assume the agent will read attachments or internal links automatically. Before merging any code, confirm with me that the changes align with the pasted context. If the agent fails to access any referenced material, stop and report the specific error rather than proceeding with assumptions.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.