Skip to content

By default, reviews run only when you ask for them or when loop-plan and loop-implement run them. The --auto option adds a block to ~/.claude/CLAUDE.md. Claude Code reads that file in every project, so it runs review-plan and review-impl by itself after plans and changes. This applies to Claude Code only. Codex does not read that file.

Terminal window
npx @tlelabs/cross-review --auto

The installer adds the block, or updates it when it is already there. Running the command again is safe.

The block is a section titled “Cross Review — automatic reviews”, between two marker lines:

<!-- cross-review-start -->
...
<!-- cross-review-end -->

It tells Claude Code to:

  • Run review-plan on a plan after it finishes writing it, before it presents the plan as final or starts coding.
  • Run review-impl on the uncommitted changes (or the branch diff) after it finishes a change and its tests pass, before it reports the work done or commits.
  • Run both without asking first, for every plan and implementation, not only those of loop-plan and loop-implement.

The block skips a review when:

  • The change is trivial: docs or comments only, or a one-line fix with no behavior risk.
  • A skill already ran the review. loop-plan gets its plan approved through review-plan, and loop-implement runs review-impl at its checkpoints.
  • You said to skip it for this change.

Only review-plan and review-impl run automatically. review-think and review-pr run only when you ask for them.

Ask Claude Code to make a small change in a project, then watch for a review-impl session on the dashboard.

The installer has no option that removes the block. Edit ~/.claude/CLAUDE.md and delete the lines from <!-- cross-review-start --> to <!-- cross-review-end -->, including both marker lines.

If the file has only one of the two markers, the installer prints a warning and leaves the file unchanged. The skills are still installed. Remove the stray marker by hand, then run the command again.