Running a review

The review screen is where you kick off a review of your working changes and watch it run from start to finish. It is the main thing you do in web mode. Before you get here, finish onboarding so a provider, API key, and model are configured, and make sure the current directory is trusted. Complete Setup returns to Home; choose Review Unstaged or Review Staged there to start the first review. If the directory is not trusted yet, the menu will refuse to start a review and point you to Settings to grant permissions first.

1. Run diffgazer inside the repository you want reviewed. It starts a local server and opens the web app in your browser, landing on the home screen.
2. Pick what to review from the menu. Choose Review Unstaged to review your working-tree changes, or Review Staged to review only what you have staged. Use the arrow keys to move through the menu, then press Enter to choose the highlighted item.
3. The selection creates a review session (POST /api/review/reviews) and takes you straight to the live progress screen for that session.

If a provider or model is not set up, the screen tells you what is missing and links you back to provider settings instead of starting. For the full provider, key, and trust walkthrough, see onboarding.

The progress screen has two panes. On the left is a progress overview of the four pipeline phases; on the right is a live activity log. Both update as the run sends events over a server-sent events stream (GET /api/review/reviews/:id/stream). The stream is the mechanism behind everything you see move on this screen, so if it stops, the screen stops updating.

The four phases run in order:

Below the phase list is the agent board. Each analysis agent gets a row showing its name, a status badge (waiting, running, done, or failed), its current action, and a progress bar. The board fills in and the bars advance as agent events arrive on the stream. If one or more agents fail, often from rate limiting, the activity pane shows a partial-analysis notice naming the agents that dropped out, and the run keeps going with what it could finish.

You can filter the activity log by agent to focus on a single one, or leave it on All to watch everything. To stop a run before it finishes, press c or use Cancel. Press Esc to return to Home without stopping the run; the server session keeps running and remains resumable from Home.

When the run itself finishes, the screen holds a short completing pause and the View Results action turns on (press Enter) — it skips the rest of that pause and takes you to the results screen, where the issues are laid out by severity and category.

• Reading results
• Review history
• The review pipeline