Issue #12

April 27, 20265 min read

Google to Invest Up to $40B in Anthropic

/config choices now persist via ~/.claude/settings.json, vim mode adds visual and visual-line selection, and --from-pr supports GitLab, Bitbucket, and GitHub Enterprise. In the news: Anthropic is investigating possible unauthorized access to its Mythos model, Google is committing up to $40B to Anthropic, and Mozilla used Mythos to fix 271 Firefox bugs.

Google Commits to Invest Up to $40 Billion in Anthropic — New York Times

On this page

Highlight
Worth Knowing
Under the Hood
From Anthropic
In the News

Highlight

Config Settings Now Persist Across Restarts

Your /config choices -- theme, editor mode, verbose output, and others -- now write to ~/.claude/settings.json and respect project, local, and policy override precedence. Settings you change once will stick across restarts, and the previously broken verbose-output persistence is fixed as part of this change.

Worth Knowing

Vim Visual and Visual-Line Modes Land in the Prompt

If you use vim mode, you can now press v for visual mode or V for visual-line mode, with selection, operators, and visual feedback. Esc in INSERT also no longer pulls a queued message back into the input -- press Esc again to interrupt.

Custom Named Themes via /theme

You can now create and switch between named themes from /theme, hand-edit JSON files in ~/.claude/themes/, or install themes shipped by plugins via a themes/ directory. /cost and /stats have also been merged into /usage, with the old names kept as shortcuts that open the relevant tab.

--from-pr Now Supports GitLab, Bitbucket, and GitHub Enterprise

--from-pr now accepts GitLab merge-request, Bitbucket pull-request, and GitHub Enterprise PR URLs, not just github.com. The new prUrlTemplate setting also lets you point the footer PR badge at a custom code-review URL, and owner/repo#N shorthand links use your git remote's host.

Hooks Can Now Invoke MCP Tools Directly

Hooks now support type: "mcp_tool", letting you call MCP tools directly from a hook instead of shelling out. PostToolUse and PostToolUseFailure hook inputs also gain a duration_ms field for tool execution time.

Native Builds Replace Glob and Grep With Embedded bfs and ugrep

On native macOS and Linux builds, the Glob and Grep tools are replaced by embedded bfs and ugrep available through the Bash tool, removing a tool round-trip on searches. Windows and npm-installed builds are unchanged.

Default Effort Raised to High on Opus 4.6 and Sonnet 4.6

If you're a Pro or Max subscriber, the default effort level on Opus 4.6 and Sonnet 4.6 is now high (was medium). Expect longer reasoning by default; override with /effort if you want to dial it back.

Under the Hood

Opus 4.7 Context Window and MCP OAuth Stability Fixes

Opus 4.7 sessions were computing /context against a 200K window instead of the model's native 1M, causing inflated percentages and premature autocompaction -- now fixed. A batch of MCP OAuth fixes also lands: missing expires_in no longer forces hourly re-auth, a macOS keychain race that caused spurious /login prompts is resolved, ${ENV_VAR} headers are now substituted, and a credential-save crash that corrupted ~/.claude/.credentials.json on Linux/Windows is fixed.

Paste, Scrollback, and Keyboard Input Fixes

Pasting CRLF content from Windows clipboards or Xcode no longer inserts blank lines between every line, multi-line paste no longer loses newlines under kitty keyboard protocol, fullscreen mode no longer snaps to the bottom on every tool finish, and Alt+K / Alt+X / Alt+^ / Alt+_ no longer freeze keyboard input.

From Anthropic

Anthropic Engineering · 7 min read

Building agents that reach production systems with MCP

A practical guide to MCP integration patterns: server design, OAuth with CIMD and vaults, context-efficient clients, and skills, plus when to reach for direct API calls or CLIs instead. Useful if you're wiring Claude Code or subagents into internal systems.

Anthropic Engineering · 7 min read

Building agents that reach production systems with MCP

Anthropic Engineering · 2 min read

Built-in memory for Claude Managed Agents

Claude Managed Agents now ship with built-in memory across tasks, users, and sessions, with no infrastructure on your side. Relevant if you're building long-running agents and have been rolling your own state layer.

Anthropic Engineering · 2 min read

Built-in memory for Claude Managed Agents

In the News

Wall Street Journal · 4 min read

Anthropic Probes Possible Unauthorized Access to Mythos AI Model

Anthropic is investigating reports that an unauthorized group gained access to its Mythos model. Worth tracking if your organization relies on Anthropic's models, as the response could shape access controls and compliance requirements going forward.

Wall Street Journal · 4 min read

Anthropic Probes Possible Unauthorized Access to Mythos AI Model

New York Times · 4 min read

Google Commits to Invest Up to $40 Billion in Anthropic

Google is committing up to $40 billion in cash and compute to Anthropic, on top of an additional $5 billion from Amazon this week. The capital and compute backing affect long-term investment in the models and infrastructure Claude Code depends on.

New York Times · 4 min read

Google Commits to Invest Up to $40 Billion in Anthropic

Wired · 5 min read

Mozilla Used Anthropic's Mythos to Find and Fix 271 Bugs in Firefox

Mozilla reports using Anthropic's Mythos model to surface and fix 271 bugs across the Firefox codebase. A concrete data point on what large-scale automated code review with frontier models looks like in production.

Wired · 5 min read

Mozilla Used Anthropic's Mythos to Find and Fix 271 Bugs in Firefox

Reuters · 3 min read

Microsoft to integrate Anthropic's Mythos into its security development program

Microsoft is integrating Mythos into its internal security development lifecycle. Notable as another major vendor using Anthropic models for security-critical engineering work.

Reuters · 3 min read

Microsoft to integrate Anthropic's Mythos into its security development program

Microsoft is integrating Mythos into its internal security development lifecycle. Notable as another major vendor using Anthropic models for security-critical engineering work.

Persistent config and merged usage views are small quality-of-life wins, but the Opus 4.7 context-window fix is the one that quietly changes behavior on long sessions.