Persistent Config, Vim Visual Mode, and Google's Forty Billion Dollar Commitment

I'm Shannon, and this is the Claude Notes Brief -- your weekly rundown of Claude Code updates and Anthropic news for the week of April twenty-seventh. Config settings now stick across restarts. Anthropic is investigating possible unauthorized access to its Mythos model. And Google is committing up to forty billion dollars to Anthropic.

Let's start with Claude Code -- the headline this week is that your config choices finally persist across restarts, which fixes a long-standing papercut. When you change your theme, editor mode, or verbose output through the config menu, those choices now write to your settings file and respect the usual project, local, and policy override precedence. The previously broken verbose-output persistence is fixed as part of the same change. Staying with the customization theme, you can now create and switch between named themes -- either through the theme menu, by hand-editing JSON files in your Claude directory, or by installing themes shipped with plugins.

The cost and stats commands have also been merged into a single usage view, with the old names kept as shortcuts that open the relevant tab. If you use vim mode in the prompt, this was a notable week -- visual and visual-line selection are now supported, with operators and visual feedback. There's also a small but welcome fix where pressing escape in insert mode no longer pulls a queued message back into the input. The pull-request workflow got broader too.

The from-pr flag now accepts GitLab merge requests, Bitbucket pull requests, and GitHub Enterprise URLs, not just github dot com. A new setting lets you point the footer PR badge at a custom code-review URL, and shorthand links now use your git remote's host. And one behavior change worth flagging for Pro and Max subscribers -- the default effort level on Opus four point six and Sonnet four point six is now high, up from medium. Expect longer reasoning by default.

You can dial it back through the effort command if you'd rather keep things shorter.

Under the hood, the fix that quietly matters most is on Opus four point seven. Sessions were computing context usage against a two hundred thousand token window instead of the model's native one million, which inflated percentages and triggered premature autocompaction on long sessions. That's now corrected. A batch of MCP OAuth fixes landed alongside it -- a missing expiration field no longer forces hourly re-authentication, a macOS keychain race that caused spurious login prompts is resolved, environment variable substitution in headers now works, and a credential-save crash that was corrupting the credentials file on Linux and Windows is fixed.

On the input side, pasting content with Windows-style line endings from a Windows clipboard or from Xcode no longer inserts blank lines between every line. Multi-line paste under the kitty keyboard protocol preserves newlines correctly, fullscreen mode no longer snaps to the bottom every time a tool finishes, and a handful of alt-key combinations that were freezing keyboard input have been unstuck.

Turning to broader Anthropic news. The Wall Street Journal is reporting that Anthropic is investigating possible unauthorized access to its Mythos model. The company is looking into reports that an unauthorized group gained access, and the response could shape access controls and compliance requirements going forward -- worth tracking if your organization relies on Anthropic's models. On the funding side, the New York Times is reporting that Google is committing up to forty billion dollars in cash and compute to Anthropic, on top of an additional five billion from Amazon this week.

That capital and compute backing affects long-term investment in the models and infrastructure Claude Code depends on. And there are two concrete data points this week on Mythos in production. Wired is reporting that Mozilla used the model to find and fix two hundred seventy-one bugs across the Firefox codebase. Separately, Reuters is reporting that Microsoft is integrating Mythos into its internal security development lifecycle.

Together, they offer a snapshot of what large-scale automated code review with frontier models looks like at major vendors. That's it for the brief. I'm Shannon, and we'll see you next week.