You open a failing GitHub Actions run and click Fix with Copilot. The cloud agent investigates the failure in its own environment, pushes a fix to your branch, reruns your CI to confirm it, and tags you for review. GitHub shipped this on May 18th for Copilot Business and Enterprise.
Where the button lives
It appears on the workflow-run logs page, on the failure itself. The agent handles what GitHub calls “simple but time-consuming work”: fixing tests, correcting linter failures. It reruns the pipeline before it tags you, so the fix arrives already validated rather than as a hopeful diff.
Since March you could trigger the same loop from a PR comment:
@copilot Fix the failing testsMay’s change moves the entry point onto the failure, so you act where you already are instead of opening a PR to ask.
Why Playwright teams feel this
A locator breaks in CI. Someone clones the branch, reproduces it locally, pushes a one-line fix, waits for the rerun. That sequence is a bounded failure with a clear scope, which is what the agent handles well. The round-trip also got shorter: cloud agent startup dropped 50% in March and another 20% in April.
The threshold shift matters more than the feature. When a CI fix costs one click instead of thirty minutes, you stop filing brittle tests under “normal maintenance.”
Turn it on, keep the gate
An administrator enables the cloud agent before the button shows up on workflow logs, and the setup docs cover the steps. The agent pushes to your branch and tags you; your review still decides what merges. Treat it as a fast first responder for bounded failures, and keep your judgment on the ones that are not.
