> ## Documentation Index
> Fetch the complete documentation index at: https://panopticon-cli.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Review Commands

> Request, restart, and abort an issue's code review

# Review Commands

`pan review` manages the review loop for an issue that already has a
workspace and a feature branch. Every subcommand calls the running dashboard,
so start it first with `pan up`. Without it, the command prints
`Dashboard not running` and exits 1.

```bash theme={null}
pan review request <id> [-m <text>]
pan review restart <id> [--model <model>] [--role <role>]
pan review abort <id>
```

Review needs no separate "start" step. `pan done` requests review as its last
step, and a PR you open or mark ready by hand starts the same pipeline through
the GitHub webhook. The verdict lands as a review on the PR, which is also
where the dashboard reads it from.

<Note>
  `pan convoy` no longer exists, and neither does `pan review run`. The
  parallel reviewers that the convoy used to launch are now the `full` review
  mode (see [Review modes](#review-modes)).
</Note>

## pan review request

Request a re-review after fixing review feedback.

```bash theme={null}
pan review request PAN-123
pan review request PAN-123 -m "Handled the null session case and added a test"
```

**Options:**

* `-m, --message <text>`: describes the fixes, and is recorded with the request

This calls `POST /api/review/:issueId/request`, the same call `pan done` makes.
The dashboard first runs verification on the branch and dispatches review once
the verified branch is pushed, so a successful request prints `Verification
started for <id>; review will start automatically when it passes`.

Depending on the issue's state, the request can instead:

* refuse, if the issue is already merged, the workspace does not exist, or the
  workspace has uncommitted changes;
* re-queue tests without a new review, if the PR is already approved but its
  checks are not green;
* succeed as a no-op, if the PR is approved and its checks are green.

**Circuit breaker:** an issue gets at most 25 automatic re-review requests.
After that the command exits 1 with `Circuit breaker triggered!`, and an
operator has to click **Review** in the dashboard to continue. The count lives
in dashboard memory, so it starts again from zero after a dashboard restart.

## pan review restart

Resume the review and re-dispatch every reviewer that has not produced its
report.

```bash theme={null}
# Resume the review with the configured models
pan review restart PAN-123

# Restart every reviewer on a different model
pan review restart PAN-123 --model claude-sonnet-5

# Re-run one reviewer lane
pan review restart PAN-123 --role security
```

**Options:**

* `--model <model>`: overrides the model for all restarted reviewers
* `--role <role>`: restarts only one reviewer lane: `correctness`, `security`,
  `performance`, or `requirements`

`restart` kills the running reviewer sessions and keeps what is already done.
Reports from reviewers that finished stay in place, and the review session is
resumed rather than started over, so a restart on the same model checks the fix
instead of re-reading the whole diff. The review session is started fresh only
when `--model` changes the model.

`--role` needs an active review run for the issue. Without one, the command
exits 1 with `Active review run not found for <id>`.

Restarting a review agent from the dashboard calls the same endpoint.

## pan review abort

Kill every running reviewer session for an issue.

```bash theme={null}
pan review abort PAN-123
```

The command prints each session it killed, and a warning if any session could
not be killed. It does not message the work agent, which stays idle. There is
nothing to reset afterwards: whether the issue is in review is read from the
PR, not from stored state.

## Review modes

The review mode is configuration, not a command flag. It comes from
`roles.review.mode` in `~/.overdeck/config.yaml`.

| Mode              | Behavior                                                                                                                                |
| ----------------- | --------------------------------------------------------------------------------------------------------------------------------------- |
| `quick` (default) | One review agent does a combined correctness, security, performance, and requirements pass.                                             |
| `full`            | Four parallel reviewer lanes (`correctness`, `security`, `performance`, `requirements`) plus a review parent that writes the synthesis. |
| `none`            | No AI review. The verification quality floor still applies.                                                                             |

```yaml theme={null}
roles:
  review:
    mode: full
    sub:
      security:
        model: claude-opus-4-8
```

Each lane's model comes from `roles.review.sub.<lane>.model`. When that is not
set, it falls back to `roles.review.model` and then to the built-in default.

## Related Commands

* [pan start](/cli/agent-commands) - Spawn a work agent
* [Specialists](/features/specialists) - The review, test, and merge specialists