You can review a change across more than one model before it ships, using a Code Review pipeline. Because Claude and Codex have different strengths, running both over the same diff catches more than either would alone. This article covers running a review, what it checks, and where the findings go.
Why more than one model
Claude Code and Codex don't catch the same things. Putting both on the same change — one may flag a correctness bug, the other a security or style issue — surfaces problems a single reviewer would miss. Because agents run on your own Claude and Codex subscriptions, using both costs you nothing extra.
Run a code review
Code Review ships as bundled pipeline presets. Start one from the global Dispatch Pipeline button, which lists your dispatch-triggered pipelines — including the Code Review presets. Pick the one you want:
- Code Review (Claude) — review with Claude.
- Code Review (Codex) — review with Codex.
- Code Review (Multi-Provider) — both weigh in, for the widest coverage.
These are starting points, not the only option — you can build your own review pipeline too. See Pipelines: overview.
What it checks
A review pass looks for correctness problems, and can also run security and compliance checks — the same kinds of checks a deep pipeline runs in its QA loop. See Anatomy of an implementation pipeline.
Where the findings go
- A review report is written to the run's output and shows up in the execution view, so you can read the whole review in one place. See Reading a pipeline execution.
- Pull-request comments — the pipeline can post its findings straight onto the PR, so the review lands where you already work.
Act on the findings
Once the comments are on the pull request, you can address them yourself — or let a hook do it. A "new PR comment" or "checks failed" hook hands the issue to an agent to investigate and fix, so a review can loop on itself until there's nothing left to flag. See Hooks: automate the busywork.
Next steps
- Pipelines: overview — how pipelines run and where to build your own.
- Reading a pipeline execution — follow the review run and read its report.
- Hooks: automate the busywork — turn review comments into automatic fixes.