Skip to main content
Mains includes a built-in code review system that lets you capture review notes and findings for each workspace. After an agent completes a run, you can request a structured code review that identifies issues, suggests improvements, and tracks review status.

​
How Reviews Work

Each workspace can have reviews attached to it. A review contains:
  • Title — a summary of what’s being reviewed
  • Status — tracks the review lifecycle
  • Findings — individual items with severity, file location, and suggestions

​
Review Status

Reviews follow a simple workflow:

​
Review Findings

Each review can contain multiple findings. A finding pinpoints a specific issue in the code:

​
Requesting a Review

There are two ways to get a code review:
  • Claude / Copilot — use the code-review tool in the agent input. The agent analyzes the diffs from the run and produces structured findings with severity ratings.
  • Codex — use Codex’s native review mode, which supports reviewing uncommitted changes, base branch diffs, specific commits, or custom ranges. Findings are parsed with priority levels (P1 = critical, P2 = warning, P3 = info).
Reviews are displayed in the Reviews section of the workspace sidebar, where you can inspect individual findings and navigate directly to the relevant code.