Claude Code Update Summary: 3rd Week of May 2026

Are you struggling to keep up with the rapid evolution of Claude Code?

New commands are added every week, settings change without notice, and procedures that worked yesterday suddenly don't. While X (Twitter) is flooded with explanations, it's hard to know what's true or relevant to you.

But don't worry. You don't need to track everything every week. By reading this article, you can catch up on all the changes that hit Claude Code this week in one go 🔥

And everything is sourced from official primary information!!

By the way, does any of this sound familiar?

• Too many weekly updates to keep track of.
• / commands increase or change names, breaking your usual workflow.
• You hear it's "gotten more convenient," but don't know how it affects your specific usage.
• Tokens seem to be disappearing faster lately, but you don't know why.
• You're frustrated not knowing if X posts are hype or official facts.

If any of these resonate with you, please keep reading.

This week alone (May 17–24, 2026), Claude Code received a series of updates from v2.1.145 to v2.1.150. Honestly, it's exhausting to track it all yourself.

This article isn't just a collection of viral posts. I've analyzed the official Anthropic CHANGELOG, release notes, and official account posts—primary information only—and reorganized them based on "how it changes things for you." Version numbers, dates, and command names are all exactly as officially stated. No hype, no omissions. You can save this and refer back to it later 👀

Source: Official CHANGELOG (github.com/anthropics/claude-code) / Official changelog (code.claude.com/docs/en/changelog) / Anthropic Official Accounts (@ClaudeDevs / @bcherny, etc.)

Let's break down this week's key points in simple terms 👇

■ 𝟭. /simplify evolved into /code-review (v2.1.147 • 5/21)

First up is a big one. A common command has been completely replaced, which might confuse you if you don't know.

The old /simplify command has been transformed into /code-review in both name and function.

The old /simplify was a "cleanup-and-fix" behavior, which has been completely removed. The new /code-review focuses entirely on being a reviewer that points out correctness bugs rather than modifying code on its own.

You can even choose the depth of the review:

1/code-review high

By specifying an effort level, you can control the intensity. Use low for a quick look and high for a serious bug hunt.

Even more powerful is the --comment option:

1/code-review --comment

This allows you to post findings directly as inline comments on a GitHub PR. You can have Claude review the code and post the results to your team's pull request in a single command.

𝗕𝗲𝗳𝗼𝗿𝗲: Use /simplify to tidy up code (but it changes things automatically).

𝗔𝗳𝘁𝗲𝗿: Use /code-review high --comment to point out correctness bugs and leave comments on the PR.

In short, the policy has shifted from "AI cleaning up" to "AI reviewing." If you used /simplify, please switch to /code-review.

■ 𝟮. Pinned background sessions no longer drop (v2.1.147 • 5/21)

This is a great improvement for those running background sessions with claude agents.

Pinned background sessions (via Ctrl+T) now behave as follows:

• They stay alive even when idle (no more automatic disappearing).
• When applying Claude Code updates, they restart and apply the update on the spot.
• When memory is tight, unpinned sessions are shed first. Pinned ones are protected until the end.

To use it, simply select a session in the claude agents list and press Ctrl+T. This ensures your most important work sessions survive. This significantly reduces the stress of background sessions vanishing during long tasks.

■ 𝟯. /usage makes token consumption transparent (v2.1.149 • 5/22)

I personally think this is the highlight of the week.

Running /usage now shows a categorized breakdown of what is consuming your limits. The units are:

• Skills
• Subagents
• Plugins
• Cost per MCP server

If you've felt like your tokens were disappearing fast, you can now identify the culprit—whether it's a heavy MCP server or a specific plugin. It's like having a detailed household budget for your AI usage.

Boris Cherny (@bcherny), the creator of Claude Code, teased this before release:

"In the next version of Claude Code: run /usage to see a breakdown of which Skills, Agents, MCPs, and Plugins are using your tokens. CLI today, coming to Desktop next" (May 21, 2026)

It's in the CLI now, with Desktop coming soon 👀

■ 𝟰. Fast mode now defaults to Opus 4.7 (5/18 • /fast)

The @ClaudeDevs account announced this on May 18:

"Fast mode now defaults to Opus 4.7 in Claude Code. Try it out today with /fast"

To clarify: Fast mode doesn't drop to a smaller model; it makes Opus output faster. You get the same intelligence with better response times. It's worth trying /fast to see the difference.

■ 𝟱. auto mode (Shift+Tab) now supports Pro plan and Sonnet 4.6 (5/22)

Two updates to auto mode announced by @ClaudeDevs on May 22:

• auto mode is now available on the Pro plan.
• Sonnet 4.6 is now supported (alongside Opus 4.7).

Auto mode is the "let Claude run" mode entered via Shift+Tab. It's now accessible to more users and models.

Additionally, v2.1.147 (5/21) fixed a behavior where auto mode would sometimes suppress necessary user questions (AskUserQuestion). Now, it correctly interprets user responses as intent signals, balancing automation with necessary confirmations.

■ 𝟲. /diff keyboard scrolling & Markdown checkboxes (v2.1.149 • 5/22)

Two small but impactful UI improvements:

First, the /diff detail view now supports keyboard scrolling. You can use:

• Arrow keys
• j / k (Vim style)
• PgUp / PgDn
• Space
• Home / End

Second, Markdown output now renders GFM task lists as checkboxes.

1- [ ] Unfinished task
2- [x] Finished task

Great for those using Claude to manage To-Do lists.

■ 𝟳. claude agents --json, PR info in status line, and plugin previews (v2.1.145 • 5/19)

Several practical additions for agent operations and plugins:

𝟭. claude agents --json: Output session lists in JSON for use in scripts (tmux-resurrect, status bars, etc.).

𝟮. GitHub info in status line: Displays the current repository and PR info in the status line.

𝟯. Plugin previews: See commands, agents, skills, and hooks before installing a plugin via /plugin.

Also, the terminal tab title now shows the count of agents waiting for input. In full-screen mode, slash commands and @mentions now support mouse hover and clicks.

For hook developers, Stop and SubagentStop hooks now include background_tasks and session_crons fields. Finally, the Read tool now returns a "PARTIAL view" instead of a hard error when a file exceeds token limits.

■ 𝟴. Security Enhancement Week (Important silent fixes)

Behind the scenes, several security and sandbox fixes were implemented.

In PowerShell, a bug allowed directory traversal (e.g., cd..) without detection, which has been fixed in v2.1.149. In Bash, a permission bypass involving environment variables was fixed in v2.1.145.

Sandbox fixes include resolving overly broad write permissions in git worktrees and a fix for a macOS crash caused by find scanning too many files. These silent fixes are the best reason to keep your version updated.

■ 𝟵. Other minor improvements

• Smarter auto-updates (v2.1.147) with better error reporting.
• Fixed duplicate history entries when using arrow keys to resend prompts.
• Pasted text now correctly reaches agents (v2.1.147).
• /feedback now includes conversation history from before context compression (v2.1.149).
• Added allowAllClaudeAiMcps for enterprise users.
• v2.1.150 (5/23) focused on internal infrastructure improvements.

Even in one week, a massive amount of progress is made. It's worth catching up once a week to stay ahead.

■ Sources (Primary Information)

• Official CHANGELOG: https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md
• Official changelog (with dates): https://code.claude.com/docs/en/changelog
• Versions: v2.1.145 (5/19) – v2.1.150 (5/23)
• Official Accounts: @ClaudeDevs, @bcherny

𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝗦𝘁𝘂𝗱𝗶𝗼 (@ClaudeCode_love) provides practical insights on Claude Code, CLI automation, and AI agent development.

We also support AI agent implementation and knowledge management through PLai Co., Ltd.

For consultations on implementing Claude Code, Codex, or Obsidian in your organization, feel free to contact us at plai-ai.com.