InnovaticInnovaticDocs
Work Modes

Review Mode

Thorough code review, autonomously. Review mode examines every change on your branch and delivers structured, actionable feedback.

Thorough code review, autonomously.

Review mode gives you a second set of eyes on your code changes — structured, detailed, and delivered without the back-and-forth of a manual review cycle.

What it does

Review mode examines all Git changes on the current branch, reads the surrounding files for context, and produces a set of structured review comments. Comments are categorized by severity — Critical, Warning, Suggestion, Nitpick — each with a precise file and line reference. The review runs autonomously without requiring interaction; when it's done, the feedback is ready to act on.

When to use it

  • Before merging a Feature or Bug agent's work into your main branch
  • As a quality gate before opening a pull request
  • When you want a second opinion on code you wrote yourself
  • After a significant refactoring to verify nothing was missed
  • As part of a team workflow to supplement human review

Run Review on the output of a Feature agent before merging. It catches what you might miss when you're focused on whether something works rather than how well it's written.

How it works

  1. Launch Review — start a Review conversation on the branch you want reviewed.
  2. Diff collection — the agent reads all changes on the branch compared to the base.
  3. Context reading — for each changed file, the agent reads the full file to understand the surrounding code.
  4. Structured comments — each finding is written as a structured comment with severity, file, line, explanation, and suggested fix.
  5. Review complete — all comments are delivered when the review finishes. No interaction required during the process.

Key features

  • Structured comments — severity levels: Critical, Warning, Suggestion, Nitpick
  • File and line references — every comment points to the exact location in the code
  • Context-aware — reads full files, not just diffs, to understand what the change means in context
  • Fully autonomous — no questions, no interaction; runs to completion and delivers results
  • Branch-scoped — reviews only the changes on the current branch, not the whole codebase

Tips

Pay attention to Critical and Warning comments first. Suggestions and Nitpicks are improvements — Critical and Warning items are what need to be addressed before merging.

You can run Review on your own changes, not just agent output. It's a useful way to catch issues before asking a human reviewer.

On this page

On this page