Playbooks · Coding agents

Review a pull request with a coding agent

A small team asks GitHub Copilot to review a pull request first, then a human still decides what merges.

Audience: Small team · Stack: A GitHub repo with pull requests, GitHub CLI installed and authenticated · Tool: GitHub Copilot code review, requested through GitHub CLI

Before you start

• A pull request that is ready for review, with a description of what it changes and why. A review agent cannot supply missing context about intent.
• Access to Copilot, and the GitHub CLI installed and authenticated — both are listed as requirements in the author code-review docs.
• Your team's review bar written down somewhere the repo can see it. The author docs say Copilot code review reads .github/copilot-instructions.md for repository-wide guidance and an AGENTS.md file in the repo root for project context, so put standing rules there rather than repeating them in every review.
• A human reviewer assigned as well. Copilot's review is advisory; see Limits.

Steps

1. When creating a pull request, request Copilot as a reviewer with gh pr create --reviewer @copilot, as the author docs show. For a pull request that already exists, use gh pr edit PR-NUMBER --add-reviewer @copilot, replacing PR-NUMBER with the pull request number.
2. Wait for the review. The author docs say this usually takes less than 30 seconds. Read the overview comment first, then the inline comments.
3. Use the severity labels on each comment — High, Medium, Low, per the author docs — to order your work, not to decide it. A Low comment can still be wrong, and a High comment can still misunderstand intent.
4. Where a comment includes a suggested change, apply it only if you agree with it after reading the surrounding code. The author docs say you can accept a single suggestion or a group of suggestions together in a single commit.
5. For a comment you want fixed by the agent, use the Fix with Copilot action on that review comment, where enabled, and check the resulting commit or pull request the same way you would check a teammate's fix.
6. If you push new changes after the review, request a re-review rather than assuming the old review still applies. The author docs note Copilot does not automatically re-review pushes unless automatic review of new pushes is configured, and that a re-review may repeat comments already dismissed or downvoted.
7. The human reviewer then does the normal review: intent, design, and edge cases the agent cannot judge. Only a human approval merges the pull request under your normal rules.

Worked example

A three-person team opens a pull request that adds rate limiting to an internal API endpoint. The author requests Copilot with gh pr create --reviewer @copilot and also assigns a teammate. Copilot flags a Medium comment: a counter that never resets, with a suggested change. The teammate reads the surrounding code, agrees, applies the suggestion, and pushes. The team then re-requests the Copilot review, confirms the comment is gone, and the teammate reviews the design question Copilot never raised — whether the limit should be per user or per API key — before approving. The agent caught a defect; the human decided the design.

What can go wrong

• Copilot repeats a comment you already dismissed. Fix: this is documented behaviour on re-review. Resolve it again and move on; do not let a repeated comment block a good change, and do not let it pass unexamined either.
• A suggested change looks right in isolation but wrong in context. Fix: do not apply suggestions blindly. Read the file, not just the suggestion.
• The review comes back empty on a large pull request. Fix: treat an empty review as no signal, not as approval. Review it yourself, and consider splitting the pull request next time.
• The team starts merging on Copilot's word alone. Fix: stop. By default Copilot leaves a Comment review, not an Approve review, and its reviews do not count toward required approvals, per the author docs.

Limits

• Copilot's review does not count toward required approvals by default, and its approval assessment alone does not satisfy merge requirements, per the author docs. Approvals from Copilot are off by default and need configuration.
• The agent reviews the diff, not your intent. It cannot tell you whether the pull request solves the right problem.
• Comments on Copilot's review comments are visible to humans but not to Copilot, per the author docs — replying to the agent in a comment thread will not get a reply.

Time and cost: NOT_MEASURED. This page gives no time or cost figure because the tool docs fetched for it state none that applies to this task.

Verification

Checked 2026-10-23 against the author's own docs. The workflow on this page was not run end to end; every command, flag, and capability claim was checked against an author doc or repo fetched that day.

• https://docs.github.com/en/copilot/how-tos/use-copilot-agents/use-code-review?tool=cli

Cut from this page, with reason:

• No claim that Copilot approves pull requests: the author docs say approvals are off by default, so the page keeps Copilot advisory-only.
• No time figure beyond the author's “usually less than 30 seconds” for the review itself; end-to-end review time is NOT_MEASURED.
• No price: the author docs require Copilot access but state no price on the fetched page, so none is given.

If a step here does not match what your installed tool shows, trust the tool's own docs for your version and stop before approving a change you cannot read.

Back to: Coding playbooks hub · Skills index

Also: Small-team hub