Claude Code Now Defaults to Manual Permission Mode
Claude CodeAnthropic changed Claude Code's default permission behavior so that "Manual" mode — which requires explicit approval before edits and shell commands — is now the default across the CLI, --help, VS Code, and JetBrains integrations. The change affects nearly every Claude Code user by default, since prior installs and fresh sessions will now stop and ask before taking consequential actions unless a different mode is explicitly configured. Alongside this, AskUserQuestion dialogs no longer auto-continue by default, giving users more control over interactive prompts, with an opt-in idle timeout available via /config.
Sources & Mentions
2 external resources covering this update
Manual Permission Mode Becomes the Default
Claude Code has shifted its baseline safety posture. As of version 2.1.200, the "default" permission mode is now labeled Manual consistently across the CLI, claude --help, and the VS Code and JetBrains extensions. In this mode, Claude Code reads files freely but pauses to ask for explicit approval before editing files or running shell commands.
Anthropic also added a manual alias so it can be referenced directly: claude --permission-mode manual or "defaultMode": "manual" in configuration now work interchangeably with the previous default value. Existing settings that reference default continue to work unchanged, but the visible label and behavior across every surface now consistently reads "Manual," reducing the ambiguity users previously ran into when comparing CLI flags, IDE menus, and documentation.
This is a meaningful default-behavior change rather than a cosmetic rename — teams and individuals who hadn't explicitly configured a permission mode will now see more approval prompts out of the box, trading a bit of friction for a safer default when running Claude Code against unfamiliar codebases or in shared environments.
Interactive Prompts No Longer Auto-Continue
Version 2.1.200 also changed how AskUserQuestion dialogs behave: they no longer auto-continue after a period of inactivity. Previously, an unanswered question dialog could silently advance, which risked Claude proceeding down a path the user never actually confirmed. Now, dialogs wait indefinitely for a response unless the user explicitly opts into an idle timeout through /config. This pairs naturally with the Manual permission mode change: both moves prioritize deliberate user confirmation over silent, time-based defaults.
Other Changes in This Release
Version 2.1.200 was largely a stabilization release outside of these two behavior changes. It fixed a startup crash when .claude.json's disabledMcpServers/enabledMcpServers fields were set to a non-array value, resolved several background-agent reliability issues (sessions silently stopping after sleep/wake, agents failing to restart after a crash left a stale daemon lock, and roster corruption disabling orphan cleanup), and improved screen-reader support by hiding decorative glyphs and giving transcript symbols short readable labels.