Review modes

EV reviews the same change in a fresh context whether you run it through a coding agent or open it in Desktop. The difference is who takes the final pass through the findings.

These paths describe how you interact with a review. They are separate from story mode, which turns a larger change into a contextual narrative inside Desktop.

Automated agent loop

Use this path when you want review to be part of the implementation loop:

1. Invoke the ev-review workflow in your coding-agent conversation.
2. The workflow starts EV with the appropriate scope and reviewer skills.
3. EV reviews the change in a separate agent context.
4. Findings return to the implementing agent as structured review output.
5. The agent addresses valid findings, records decisions, and reviews again when needed.

Agent	Invocation
Codex	$ev-review
Claude Code	/ev-review

See Run your first review to install the optional workflow.

With Desktop

Use Desktop when you want to read the review and make the final decisions yourself:

ev review -b origin/main --desktop

1. Inspect the changed code, summary, findings, and attached evidence.
2. Resolve findings you have addressed, or dismiss them with a reason.
3. Add comments or evidence where the automated review missed context.
4. Choose Submit review.

The waiting terminal receives a Markdown handoff. Desktop also copies it to the clipboard so you can return it to a coding agent.

Within Desktop, switch between Diff for a file-by-file view and Story for a narrated walkthrough of a broader change. Both views belong to the same review.

Choose a path

Choose…	When…
Automated agent loop	You want the agent to review, fix, and repeat without a manual handoff.
Desktop	You want to inspect the evidence, add your own feedback, or decide how each finding should be handled.

You can switch paths between review passes. Findings, evidence, comments, and decisions remain part of the review rather than belonging to one interface.