Updates in Claude Code: New Depth Levels for Code Review
Claude Code has introduced new depth levels for its code review tool, allowing users to choose the optimal approach for code checking.
In /code-review from Claude Code, depth levels of analysis are now available, and the review process has been reworked for each level.
At the low level, /code-review identifies more issues than other code review tools, with significantly lower token costs. The high level provides a noticeably higher recall when a deeper analysis is needed. The compromise is yours to choose.
Previously, /code-review used a single fixed prompt regardless of the chosen depth level. Now, a separate review strategy is applied for each level. The team automatically uses the depth level set for the current session, or it can be specified explicitly, for example: /code-review high.
At the low level, a single pass over the diff is performed. This is fast and cheap enough to run checks before every push.
At the medium level, the modified code is analyzed in context, followed by several passes searching for issues from different angles, after which each identified issue is verified before it makes it to the final report.
At the high level, search and verification agents are launched as sub-agents with clean context, so they are not tied to the reasoning of the agent that just wrote the code. At the x-high level, the review goes even deeper and covers the potential impact of changes on code beyond the diff itself.
There is one more level above high: /code-review ultra. It launches a whole group of review agents and independently reproduces each identified issue. The coverage of critical issues remains the same as at the high level, but false positives are significantly reduced. At Anthropic, this level is run for every PR.
Why it matters
AnalysisThis update allows developers to review code more effectively, reducing costs and enhancing the quality of analysis. The flexibility in choosing the depth level of analysis opens new opportunities for optimizing the code review process.
Discuss in community
Share your questions and insights with developers
Related in Ecosystem
Practical guides, core terminology, and agent skills
Claude Code Guide: Setup for Agentic Programming (Beginner-Friendly)
A comprehensive practical guide to setting up Claude Code: CLI installation, CLAUDE.md and settings.json configuration, permission matrices, automated hooks, custom /truth command, and subagents.
Verification Fatigue & Rubber-Stamping
Cognitive dulling of an engineer's attention due to a constant influx of large diffs generated by AI, leading to mechanical approval of unverified code in production.
Claude Code Projects: How a Single AI Orchestrates a Team of Autonomous Agents
A deep dive into Claude Code Projects: Coordinator and Threads architecture, parallel branches, shared memory via MEMORY.md, and cloud agent management.
AI as Code Reviewer (Bug Detection Before Release)
A methodology for utilizing language models as a stringent senior engineer for automated code audits (Code Review). It identifies hidden security vulnerabilities, memory leaks, and architectural bugs before the software rollout.