← Blog
October Changelog
Macroscope
Macroscope
Product
Changelog

October Changelog

We shipped a bunch of improvements to Macroscope this month. Here's an overview.

Here's what we shipped this month.

Smarter Approvability & new modes

Approvability decides whether a PR is safe to merge without a human, and approves it if so. It now runs on an upgraded model that handles longer, more complex policies more reliably.

The new Approvability Mode sets how cautious it is:

  • Permissive (default): Relies on the PR's verifiable evidence, like Correctness, Check Run Agents, CI, review comments and CODEOWNERS, and saves human review for material risk.
  • Strict: Closer to the previous behavior. It asks for human review when meaningful risk or uncertainty remains.

Your .macroscope/approvability.md rules apply in both modes. Set the mode per repo under Settings → Repos → Approvability. Learn more about Approvability Mode.

Approvability Dashboard

See what Approvability approved, what it didn't, and why, in Code Review → Approvability. Filter by repo, date range or verdict, search PRs, and export to CSV. Learn more about Approvability.

Preview and Run Modes for Approvability

Preview Mode shows which PRs Approvability would approve, without submitting real approvals. Admins can turn it on per repo, in bulk, or as a default in Settings → Approvability. Learn more about Preview Mode.

Run modes, set per repo under Settings → Repos → Features, are Always run, Manual invocation only (via @macroscope-app or an Always Review PR Label), and Never run. Repos that had Approvability on are now Always run, and repos that had it off are now Never run. Learn more about run modes.

Policy Repo

Write your code review config once and share it across your repos. Pick a Policy Repo under Settings → Repos → Policy Inheritance, and linked repos inherit its .macroscope/ files. A file in a repo's own .macroscope/ folder overrides the inherited one. Learn more about Policy Inheritance.

Faster, More Accurate Code Review

Correctness reviews are now 14% faster on average, with 5% higher recall on known bugs (MacroscopeBench) and no change in precision. We removed the step that walked the code's AST for extra context, and made other optimizations, so the agents run more efficiently and find more bugs.

The Little Things

  • "onlyApprove" for Approvability: A front matter option in approvability.md. Approvability skips any PR that touches paths outside the patterns you list.
  • Auto Rerun for Approvability: Resolve the last blocking thread and Approvability reruns, so a fixed PR gets approved without a new commit.
  • Check Run Agents follow default ignore rules: In repos without a custom ignore.md, they skip the same files Correctness skips. Use the "include" front matter to review ignored files.
  • "ignoreTests" for Check Run Agents: Set ignoreTests: false to have one agent review test files.
  • New Check Run Agent models: ember-1 (built on Kimi K3, 1M-token context), claude-sonnet-5-5 and gpt-6-1-sol.
  • Fix It For Me PRs request reviewers: You're requested automatically, plus the original author if the fix is stacked on their branch.
  • See when an agent's instructions changed: In the Code Review activity log, hover a Check Run Agent's name to see the commit that last changed its file.
  • New default repo settings: Skip PRs by Author, Labels or Target Branches, and Always Review PR Labels, are now under Settings → Repos → Defaults for newly linked repos.

Loving Macroscope? Tag us @Macroscope on X, email contact@macroscope.com, or reach out on Slack. We'd love to hear your feedback!

Haven't tried Macroscope yet?
Get started with $100 in free usage.