Old Copilot Workspace tutorials describe a GitHub Next experiment. Its official manual is archived; do not expect today’s issues to have the former “Open in Workspace” interface. For current issue-driven development, GitHub documents assigning a task to Copilot cloud agent and reviewing its resulting pull request.
Official documentation checked October 14, 2026. The steps below concern current cloud-agent issue assignment, not the historical Workspace preview.
Prepare an Issue That Can Be Verified
Choose a small bug or feature in a repository you are authorized to change. Write the observed behavior, expected behavior, reproduction steps and acceptance tests. Identify the relevant code when you know it, but do not invent an implementation merely to make the issue look complete.
A useful issue might say: “The product list accepts a negative page number and returns an internal error. Return the API’s standard validation error instead. Preserve existing behavior for omitted pagination. Add tests for negative, zero and valid page values. Do not change authentication or dependencies.” This gives the reviewer concrete conditions to verify.
Confirm plan access, repository permissions and organization policy. Cloud-agent tasks can modify code and create branches or pull requests; assigning the issue is authorization to start work, not just to display a plan. If you need an approval checkpoint before any implementation, create and review a plan first in an appropriate editor workflow such as VS Code Plan mode.
Assign the Issue and Review the Resulting Work
- Open the intended repository on GitHub, then open the issue from its Issues tab.
- Use Assignees and select Copilot when offered. GitHub marks this issue-assignment workflow as preview, so consult its current instructions if the interface differs.
- Review the target repository and starting branch. Add scoped instructions, including files that must not change and required tests. Do not accidentally send a private issue into a public repository.
- Submit the assignment and follow the session or pull-request link. Read what the agent actually changed rather than relying only on its summary.
- Check the diff against every acceptance criterion. Look for unrelated edits, weakened tests, new dependencies, secrets and changes to CI configuration.
- Review validation evidence and any request to run GitHub Actions. By default, workflows on agent changes can require human approval; do not assume every check has already run.
GitHub’s documented flow carries the issue’s existing context into the task. Later issue comments are not automatically a way to steer the in-progress agent; put follow-up instructions in the resulting pull request as directed by GitHub. Be explicit about whether you want an explanation or another code change.
Handle Failures Without Expanding the Task
Copilot is not available as an assignee: check eligibility, write access and repository policy. Do not look for a hidden Workspace beta switch or buy a subscription until the missing requirement is known.
The implementation is incomplete: point to the failing acceptance criterion and actual test output. Ask for the smallest correction. If the issue was ambiguous, resolve the requirement before allowing further edits.
Tests did not run: inspect the session and CI logs. Missing dependencies, unavailable credentials and policy can prevent verification. Run the relevant checks in a controlled environment and report what remains untested; a passing-looking summary is not a substitute.
The task changes unrelated files: request a narrowed diff and review again. Preserve your repository’s branch protections and required reviews. Do not bypass them merely because the author is an AI agent.
Finish by linking the reviewed pull request to the issue and confirming the behavior in the environment that matters. Keep planning, implementation, test success and merge approval as separate checkpoints.