# PR#104: The Codex Review Holds the Merge on werkrevisor

> **WARNING:** This document describes only the change applied in this PR.
> It may already be outdated once the next PR is merged.
> Historic PR documentation is not maintained along with new PRs; treat it as a snapshot.

## Related Links

- [PR#88](2026-10-11-PR#88-codex-runner.md): the runner and `[codex] mode`.
- [PR#89](2026-10-11-PR#89-codex-advise-werkrevisor.md): `advise` on werkrevisor.

## Problem

werkrevisor is the merge gate of all werk repositories and reviews its own PRs.
A change that weakens the gate is then reviewed by the weakened gate.
The Codex review is independent of werkrevisor;
in `advise` it found real defects that werkrevisor missed, e.g. in PR#90 and PR#91.
The user chose werkrevisor as the first repository with `gate`.

## Non-Goals

- No `gate` in other repositories; werkator stays on `advise`.
- No change of the runner.

## Solution

- `.werkrevisor.toml`: `[codex] mode = "gate"`.
- Werkbaum: `#model.codex.gated`, `gate` in werkrevisor.
- Version 0.41.3.

The runner reads the mode at the head of `main`, so this PR itself is not held by `ai-review/codex`.

Steps of the human after the merge, in this order:

1. Wait until the runner has set `ai-review/codex` on the open PRs, at most 5 minutes.
2. Add `ai-review/codex` to the required status checks of `main`, e.g. with
   `tools/branch-protection https://git.javagil.de/mi/werkrevisor main --contexts ai-review/codex --approvers ai-reviewer,mi`,
   first with `--dry-run`: the tool keeps the required contexts, but also sets its other fields.

With the workstation off, a werkrevisor PR waits with the status `pending`.

## Open Questions

None.