Purpose
Get automated pull request review running on the repositories that benefit from it, at a depth that matches each codebase, without making developers configure anything.
When to use
- An organization on a Claude Team or Enterprise plan is enabling Code Review for the first time.
- New repositories need to be added to an existing Code Review setup.
- Review depth needs revisiting: routine repositories feel over-reviewed, or a critical one is not getting a deep enough pass.
- Pull request volume has grown faster than human review capacity — the usual effect of agent-assisted development.
Instructions
Confirm eligibility. Code Review is generally available for Claude Team and Enterprise plans. Check the credit program before rolling out, since it is time-bound: see references/review-depth-levels.md.
Enable and connect (admin steps).
- Enable Code Review in your Claude Code settings.
- Install the GitHub App.
- Select the repositories you want reviews to run on.
Choose a depth per repository. Admins set Quick, Standard, or Deep for each repository. Standard is the default; every level delivers high-quality reviews, so the choice is about how much review each codebase warrants, not about settling for less.
- Routine codebases → a quick pass.
- The most critical codebases → a deep review.
- Everything else → Standard. The per-level description and selection criteria are in references/review-depth-levels.md.
Tell developers there is nothing to do. Once Code Review is enabled for a repository, reviews run automatically on new pull requests with no developer configuration.
Set expectations on turnaround. Most reviews finish in about 6–10 minutes, roughly 3x faster than during the research preview (~20 minutes). Plan review-blocking checks around that window rather than around preview-era timings.
Read findings as problem / trigger / fix. Findings are written to state the problem, what triggers it, and the fix in a few sentences. Treat anything that does not fit that shape as a signal to re-read the diff rather than as a verdict.
Revisit depth with evidence. If reviews feel noisy on a routine repository or thin on a critical one, change that repository's depth. Depth is a per-repository setting precisely so one organization-wide choice does not have to fit every codebase.
Examples
- First rollout on a Team plan. An admin enables Code Review, installs the GitHub App, and selects three repositories: the main product service (Standard), a payments service (Deep), and an internal tooling repository (Quick). Developers are told only that reviews will appear on new PRs.
- Reacting to review noise. A repository of generated client bindings produces long reviews on mechanical changes. The admin moves that repository from Standard to Quick; the critical services keep Standard and Deep.
- Raising depth for a risky codebase. An authentication service moves from Standard to Deep ahead of a migration, so reviews cover it more thoroughly while the risky changes land.
- Claiming credits. An eligible Enterprise organization claims its Code Review credits before January 15, 2027, and schedules the rollout so the repositories are already enabled when the 30-day credit window starts — see references/review-depth-levels.md.
Notes
- Code Review is designed to catch bugs before they reach production; it is one input to a human decision, not an approval.
- Anthropic runs Code Review on its own pull requests, which is part of how the team finds where reviews slow engineers down.
- Consult the official Code Review docs for the settings and GitHub App details; this skill covers the rollout decisions, not the UI.
Source
Distilled from Code Review is now generally available (published 2026-10-15).