Claude Code makes code review more explicit

When Claude moves fast, it is tempting to focus on speed. Today's useful signal is less flashy: review. Claude Code 2.1.146 makes the review command clearer, lets auto mode ask questions when the workflow depends on them, and fixes MCP lists that could miss resources after the first page. For a team letting Claude touch real projects, that kind of small release can prevent a surprising amount of confusion later.

What actually changed in Claude Code 2.1.146

Claude Code 2.1.146 was published on May 21, 2026. The most visible change is that /simplify has been renamed to /code-review, with an optional effort level such as /code-review high. That may sound cosmetic, but the name is much more honest: ask Claude to review a change, not merely make something simpler.

The release also includes three practical fixes for everyday work:

• Auto mode no longer suppresses AskUserQuestion when the user or a skill explicitly relies on it. Auto mode means Claude Code can push work forward with less handholding, but some decisions still need a human answer.
• MCP lists for resources/list, resources/templates/list, and prompts/list no longer drop items after page one on paginating servers. MCP, the Model Context Protocol, is the standard that connects Claude to tools, resources, and reusable workflows outside the model.
• Managed settings for forceLoginOrgUUID and forceLoginMethod should now be enforced for third-party provider and API-key sessions. For teams, login rules become harder to bypass by accident.

Source: Claude Code changelog plus GitHub release v2.1.146 and npm registry for @anthropic-ai/claude-code.

Why this matters for practical teams

A coding agent is an AI assistant that can read code, suggest edits, run commands, and sometimes work through several steps without you answering every line. It becomes useful when the team knows where speed ends and control begins.

Picture a consultant with Claude Code connected to a client project. Claude can help clean up a booking flow, write tests, and spot risks. Before anything moves forward, the human still needs three answers: what changed, what Claude assumed, and which questions were too important to guess.

That is why this release is worth noticing. /code-review high gives teams clearer language for review. AskUserQuestion in auto mode makes it more reasonable to build skills that pause when a business decision is missing. Fixed MCP lists reduce the chance that Claude only sees the first page of your prompts or resources and misses the instruction that actually applies.

Source: Claude Code security docs plus Claude Code settings docs and MCP architecture overview.

What to test today

Choose one change that is already in motion, such as an internal admin view, a proposal flow, or a small customer portal. Do not ask Claude to build more right away. Ask it to review the change as if it had to explain the work to a colleague who was not in the session.

Run this in Claude Code on a branch or copy where you can compare the result. Pass secrets through environment variables or a secret manager, use scoped tokens, keep file-write permissions narrow, and make uncertain changes require approval before tests, commits, or deployment.

If you already use Claude Code across several projects, this is a good Tool Forge track: standardize how review commands, MCP resources, permissions, and review notes should work, so every project does not invent its own rules.

Try this prompt this week

Use this in Claude Code after updating to the latest version. Adjust the bracketed fields first.

You are the review lead for a Claude Code task.

Project: [short description of the repo, product, or customer flow]
Change to review: [branch, diff, folder, or pull request]
Important business context: [what must not break for the customer, user, or team]
Relevant MCP resources/prompts: [list servers or write "inventory first"]
Actions you may not take without approval: [commit, deploy, change database, change customer-facing text, run destructive commands]

Do this before suggesting code changes:
1. Create a review plan and say whether `/code-review high` is justified or whether normal effort is enough.
2. Inventory the MCP resources, templates, and prompts you plan to use. Say explicitly if any list looks incomplete.
3. Use `AskUserQuestion` when you are missing a decision that affects users, pricing, legal, security, or brand. Do not guess.
4. Split findings into: must fix, should discuss, can wait.
5. Suggest tests or manual checks a human can run in under 20 minutes.
6. Finish with a short review note for the pull request: what you checked, what you did not check, and what the human must approve.

A good result looks like this:

• Claude asks for missing decisions instead of pretending to know.
• The review points to concrete risk, not only style.
• MCP resources and prompts are visible in the work note.
• The human can approve, pause, or reject the change without reading the whole conversation.
• Secrets, customer text, and raw data do not land in logs unnecessarily.

What to watch next

I would keep an eye on two things. First, whether /code-review becomes a clearer hub for repeatable team routines, not just a new slash-command name. Second, whether the MCP fixes make agent workflows more reliable when teams have many prompts, resources, and templates connected to the same workspace.

This is not a glamorous release. For many small teams, that is the point. It makes Claude a little easier to review before it affects something important.