Copilot cloud agent is now a scriptable platform: REST API, SDK GA, and scheduled tasks

A cluster of early-June GitHub changelog entries adds up to one shift: Copilot's cloud agent is now an automation surface you can script, not just a button in the UI.

The Agent tasks REST API (public preview, June 4) lets Pro, Pro+, and Max users programmatically start and track cloud agent tasks — the agent works in its own dev environment, validates changes, and opens a PR. GitHub's own suggested uses are the interesting part: fan out a refactor or migration across many repositories from a script, bootstrap repos from an internal developer portal, or auto-prepare a weekly release with notes. Auth works with classic/fine-grained PATs and OAuth tokens.

Around it: the Copilot SDK went GA (June 2) for building Copilot-powered tools, scheduled and automated cloud agent tasks landed the same day, and "Fix with Copilot" for failing Actions jobs plus larger context windows with configurable reasoning levels arrived June 4.

If your team is on Copilot, the comparison worth making this week is against the Claude Code SDK and Cursor SDK: all three vendors now expose headless, scriptable agents, and the differences are in pricing model (Copilot's new AI Credits meter these API-started tasks), repo integration, and how much control you get over review gates. The cross-repo migration script that was a custom harness project in 2025 is now a curl call.

Sources: GitHub Changelog: Agent tasks REST API, Copilot SDK GA, Scheduled agent tasks, Fix failing Actions