Claude Code Stops Asking Permission For Everything

Claude Code

What's new from Anthropic's AI coding assistant

🔄 Improved · Permissions Claude Code gets smarter about when to ask for permission The tool is now better at auto-approving safe filesystem commands (like LANG=C rm foo or timeout 5 mkdir out) in Accept Edits mode. It also automatically approves sandbox network access in auto and bypass-permissions modes, so you'll see fewer interruptions for routine operations.

🛠️ New · Focus View
New focus toggle helps you see what matters most Press Ctrl+O in no-flicker mode to get a clean view showing just the prompt, a one-line summary of what tools Claude used (with edit stats), and the final response. Perfect when you want to cut through the noise and see the essentials.

⚡ New · Status Line Your status bar can now refresh itself automatically Added a refreshInterval setting that lets your status line command re-run every N seconds. Great for showing live git status, build progress, or other info that changes over time.

🔧 Fixed · Bugs Major fixes for permissions, MCP connections, and more Fixed the --dangerously-skip-permissions flag getting downgraded after approving protected writes · Fixed MCP HTTP connections leaking ~50MB per hour when servers reconnect · Fixed managed settings not updating until restart · Fixed several /resume picker issues including empty lists and missing diffs.

🔧 Fixed · Bugs Better handling for non-English text and various UI issues Korean, Japanese, and Unicode text no longer gets garbled when copied in no-flicker mode on Windows · Fixed slow mouse scrolling on Windows Terminal · Fixed keyboard shortcuts not working in Warp terminal · Fixed memory leak during API retries.

The AI ChangeLog is generated daily by Demand Signals. We scrape official changelogs, run them through Claude, and publish a plain-English summary so you don't have to read the docs. Subscribe to our blog for daily updates.