GitHub Copilot coding agent: model picker, self-review and built-in security scanning

In one sentence GitHub rounds up what is new in the Copilot agent: per-task model choice, self-review with Copilot code review before opening the PR, code/secret/dependency scanning in the workflow, custom agents in .github/agents/, and session handoff with the Copilot CLI, which went GA the day before.

Verified Official source

ShareLinkedIn X

In May 2025 GitHub launched the Copilot coding agent: you assign it a task, it works in the background on a branch and opens a pull request.

On February 26, 2026 the GitHub Blog rounds up what has landed:

Model picker: in the Agents panel you pick the model per task, a faster one for routine work, a stronger one for hard tasks, or "auto". Available for Copilot Pro and Pro+, with Business and Enterprise coming.
Self-review: before opening the PR the agent runs Copilot code review on its own changes and fixes the patch; the task logs show where it did so.
Security in the workflow: code scanning, secret scanning and vulnerable-dependency checks run inside the agent's work, and problems such as a committed API key are flagged before the PR opens.
Custom agents: you create a file under .github/agents/ in the repo to define a specific approach, shareable across an organization or enterprise.
CLI handoff: from the cloud you continue locally with "Continue in Copilot CLI"; from the CLI you delegate back to the cloud by pressing &.

The day before, on February 25, the Copilot CLI became generally available to all Copilot subscribers.