Claude Code v2.1.221 adds Focus view, file masking, and permission fixes
The update introduces a Focus view toggle for cleaner chat interfaces, a 'mask' mode for sandboxed credential files on Linux/WSL, and resolves several permission and authentication issues across platforms.
1 source · cross-referenced
- Adds Focus view toggle (Ctrl+Alt+F) to hide tool activity behind expandable summaries in VS Code
- Introduces 'mask' mode for sandboxed credential files on Linux and WSL to substitute real values on egress
- Fixes permission bypass in Bash tool for zsh regex conditionals and PowerShell path handling on Windows
- Resolves authentication failures for Bedrock with AWS SSO named profiles on Windows with stray HOME variable
- Improves auto mode permission checks, prompt auditing, and session management across platforms
Claude Code v2.1.221, released by Anthropic, introduces a Focus view toggle for VS Code that collapses tool activity into an expandable per-turn summary with a live running-tool indicator, accessible via Ctrl+Alt+F or the "Claude Code: Toggle Focus view" command. The update also adds a 'mask' mode for sandboxed credential files on Linux and WSL, where sandboxed commands read a sentinel copy of the file while a sandbox proxy substitutes the real values on egress. On macOS, file masking falls back to a deny mechanism.
The release resolves a Bash tool permission bypass affecting zsh users, where hidden commands in regex conditionals (e.g., [[ ]]) could execute without prompting. PowerShell permission checks are also fixed for paths containing quote characters on Windows, ensuring proper approval prompts. A separate issue where disabling thinking mode persisted incorrectly for the rest of a session has been corrected.
Authentication issues with Bedrock using AWS SSO named profiles on Windows machines with a stray HOME environment variable are addressed, preventing failures in desktop-managed sessions. The claude-api skill gains a prompt-audit subcommand for reviewing prompts and tool descriptions for patterns written for older models.
Auto mode permission checks are made more efficient by reusing cached conversation prefixes, and switching modes while a check is pending now reliably prompts for permission instead of applying stale results. Prompt-cache costs for these checks are reduced, and the Stats panel is updated to include cache token breakdowns.
Session management improvements include honoring falsy values for CLAUDE_CODE_RESUME_INTERRUPTED_TURN to disable auto-resume, preventing race conditions during wake-from-sleep that could force re-authentication, and ensuring renamed sessions update consistently across CLI and desktop interfaces.
Plugin and skill handling is refined: plugins installed via /plugin now activate immediately when safe, and the root-level SKILL.md validation error suggests using the plugin root. Sessions forked with /fork now create a new worktree instead of working in the original session's directory.
- Aug 4, 2026 · TechCrunch — AI
AWS embeds Superblocks’ vibe-coding tool in enterprise private clouds
Trust79 - Aug 3, 2026 · TechCrunch — AI
House spending records show ChatGPT dominates paid AI use on Capitol Hill
Trust76 - Aug 2, 2026 · AWS — Machine Learning Blog
AWS details Amazon Bedrock AgentCore Observability for diagnosing production agent performance
Trust79