Best practices

Review with a different model

A second model catches what the first one missed. Make the reviewer run on a different CLI and model from the agent that wrote the commit.

reviewer:
  runtime: ["claude:claude-opus-5-5:high", "codex:gpt-6.1-sol:high"]
  different-runtime-from: implementer

How it picks

The reviewer uses the first runtime in its list that differs from the implementer's recorded one. A different effort alone does not count.

Keep both installed

If no different runtime is available, the reviewer does not run, rather than review its own model's work.