# Quickstart

> Make a change, ask for a review with review-impl, follow the rounds while they run, read the outcome and open the dashboard.

This walkthrough reviews one uncommitted change in an invented project, `inkwell`, a notes app. It assumes cross-review is installed and `codex` is logged in; see [Install](https://cross-review.tlelabs.com/docs/start/installation/).

## 1. Make a change

Add a Markdown export to `inkwell`, with a test. Leave the changes uncommitted:

- `src/export/markdown.js`: a new function, `exportNoteToMarkdown`.
- `test/export.test.js`: a test that checks the heading and the body of an exported note.

Run the project’s tests before you ask for a review, so the reviewer sees the change in its final state.

## 2. Ask for a review

Ask the agent to review the change with `review-impl`. Say what the change should do, because the reviewer checks the change against the request.

**Claude Code**

```text
/review-impl Check my uncommitted changes. The request: add Markdown export for a note, with a heading for the title and the body unchanged.
```

**Codex**

```text
$review-impl Check my uncommitted changes. The request: add Markdown export for a note, with a heading for the title and the body unchanged.
```

The skill works out the review mode from git: uncommitted changes make it a working-tree review. You do not pick a mode, a model or an effort unless you want to. Without a choice, the reviewer uses the settings in [Settings](https://cross-review.tlelabs.com/docs/guides/settings/) or its own defaults.

## 3. Follow round 1

The agent starts the review and waits for the reviewer. While it waits, it reports what the reviewer is doing, for example that it is reading `src/export/markdown.js`. The reviewer runs in the background, so a review can take several minutes.

The agent stops waiting when the round ends and acts on the result. You do not need to do anything until it reports.

## 4. Read the outcome

Each round ends with one of these results. The agent acts on it and reports to you.

| Result     | Meaning                                                     | What the agent does next                                                                                                       |
| ---------- | ----------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------ |
| `respond`  | The reviewer raised issues                                  | Fixes the accepted ones, disputes the rest with evidence, and sends the next round                                             |
| `done`     | The reviewer approved, or the two sides reached a stalemate | Ends the session and reports the verdict; a stalemate is yours to decide                                                       |
| `ask_user` | The round ended without a verdict                           | Shows you the reviewer’s text and asks how to go on                                                                            |
| `stopped`  | An error, a stall or a timeout ended the round              | Stops the session and reports the reason; see [Troubleshooting](https://cross-review.tlelabs.com/docs/guides/troubleshooting/) |

A round with issues looks like this. The numbers and names are invented:

```markdown
### ISSUE-1: Title line is written twice
- Category: bug
- Severity: medium
- Location: `src/export/markdown.js:18-22`
- Problem: the heading is added twice when the note has an empty body.
- Evidence: `if (!body) lines.push(heading)` runs after `lines.push(heading)`.
- Why it matters: the exported file has two identical headings.
- Suggested fix: return after the first push, and add a test for an empty body.


### VERDICT
- Status: REVISE
- Reason: one defect in the empty-body path.
```

The agent fixes `ISSUE-1`, adds the test, and replies with what it fixed. The reviewer checks the fix in the next round. Issue numbers stay the same from round to round. [Core concepts](https://cross-review.tlelabs.com/docs/start/concepts/) explains how issues, verdicts and disputes work.

> **Tip:**
>
> You can say what to do in chat at any point. For example, “accept ISSUE-2 as it is” or “stop the review”. The agent follows your latest instruction over the default steps.

## 5. Open the dashboard

The dashboard shows every session in the folder and the rounds of each one, with the issues and their state. Start it from the repository:

```sh
cross-review dashboard
```

The command prints a link and opens it in your browser. The page runs only on your machine. See [Dashboard](https://cross-review.tlelabs.com/docs/guides/dashboard/) for its views and options.

## Next

- [Core concepts](https://cross-review.tlelabs.com/docs/start/concepts/): sessions, rounds, issues and verdicts.
- [review-impl](https://cross-review.tlelabs.com/docs/skills/review-impl/): the skill in full, including branch and commit reviews.
- [Plans](https://cross-review.tlelabs.com/docs/guides/plans/): for larger work, plan first with `loop-plan`.