Anthropic S-1 Shows Red-Hot Growth

Highlight

Guided auto mode setup reaches every backend

/auto-mode-setup walks through your project and recent sessions to teach auto mode what your environment looks like, so it prompts you less on the commands you actually run. It now works on Bedrock, Vertex, Foundry, behind the Claude apps gateway, and in sessions with telemetry disabled, which previously had no path to it. Auto mode's prompt before the first read outside your working directories also now explains that blocking those reads makes shell commands ask for permission.

Worth Knowing

Code Review gains depth levels and a stop command

Owners can now set Quick, Standard or Deep as the organization-wide review level and override it per repository, plus opt in every repository in a GitHub organization at once, including ones added later. You can also end an @claude review always subscription on a pull request with @claude review stop without kicking off another review.

sed -i now runs as written instead of previewing

sed -i no longer goes through the simulated file-edit preview; it gets the standard Bash permission prompt and runs exactly as written. Note that, like other Bash edits, those changes are not restored by /rewind.

VS Code collapses background subagents to one line

Background subagents no longer stream every step into your chat panel; they show as a single status line under their Agent card, with the full transcript still available in /tasks. Edit and Write in the panel also got faster, with the wait for new problems after each edit dropping to 0.75 seconds from 1 to 1.5 seconds.

Resuming an old conversation now warns about usage

If you are on Pro or Max and resume an inactive conversation that would eat a noticeable share of your 5-hour limit, Claude Code now says so before you spend it. You can resume anyway, or start fresh and let Claude look up the previous conversation if you refer to it.

Usage warnings become one-line fixes you can apply

Each warning in /usage is now a single actionable line that names where it came from, and pressing f applies the fixes. The same suggestions show up in VS Code and /doctor.

Ultrareview now requires workspace trust

claude ultrareview asks you to confirm workspace trust in a terminal and refuses to run without it elsewhere. In scripts, pass --trust-workspace (it takes no value, so put a PR number or base branch before it) or set CLAUDE_CODE_DISABLE_ULTRAREVIEW_TRUST_CHECK=1.

Under the Hood

Sandbox and permission gaps closed across platforms

Sandboxed Bash commands no longer run your shell startup files outside the sandbox, denyWrite globs now apply through symlinks, sandboxed and auto-accepted edits can no longer modify the file passed to --mcp-config, and Bash and PowerShell permission checks now prompt for compound cd commands and redirects they previously waved through.

Compaction loops and memory use improved

Sessions no longer compact repeatedly after reading a very large nested CLAUDE.md, compaction after a "Prompt is too long" error trims dense content like base64 and SVG properly, and non-interactive runs (-p, Agent SDK) use less memory when compacting very large transcripts.

In the News