GitHub Copilot code review drops its 20,000-line PR cap
The 300-file cap is gone, bot-opened PRs get a full agentic review, comments carry a resolution reason — plus Claude Code 2.1.250 and Cursor in the AI SDK.
Copy markdown
The 300-file / 20K-line cap is gone
Copilot code review used to refuse any pull request over 300 files or 20,000 lines. That ceiling no longer applies, so the big refactors, framework bumps, and vendored-dependency PRs you used to review by hand finally get a real agentic pass.
Bot-opened PRs get a full review
Reviews auto-requested on PRs opened by bots — including the Copilot cloud agent — now get the full agentic review (billed to your org, with the no-license policy enabled) instead of the old degraded fallback. If you run agents that push PRs, their work now gets checked like anyone else's.
Resolve a comment as 'Won't fix'
You can now close a Copilot review comment with a reason — Addressed, Won’t fix, or Incorrect. It clears the thread and feeds signal back to trim false positives on future reviews.
Elsewhere: Claude Code 2.1.250
Anthropic shipped a reliability and bug-fix point release today, on the heels of 2.1.248’s --restricted mode for untrusted repos. Update to pick up the latest session and prompt-cache fixes.
Elsewhere: Cursor lands in the AI SDK
Vercel's new @ai-sdk/harness-cursor adapter lets you drive Cursor's coding agent through the same HarnessAgent interface as its other harnesses — one code path to swap agents behind your app.