Claude Code 2.1.200 Renames the Default Permission Mode to "Manual" and Changes How AskUserQuestion Works

Claude Code 2.1.200 released July 3 with two changes that affect how sessions behave when Claude needs your attention, plus a large set of background agent reliability fixes.

”Default” is now “Manual”

The permission mode previously shown as “Default” has been renamed to “Manual” across the CLI, --help text, VS Code, and JetBrains. The behavior is the same: Claude pauses and prompts you before taking any action that needs approval. Only the label changed.

The rename is a clarification. “Default” as a label implied you were running the standard configuration, which confused users trying to understand the differences between permission modes. “Manual” is more direct about what actually happens: you review and approve things manually. The rename also better distinguishes this mode from Auto mode, which runs background safety checks instead of prompting.

AskUserQuestion dialogs no longer auto-continue

Previously, when Claude opened an AskUserQuestion dialog, it would automatically continue after an idle timeout if you hadn’t responded. That default is off now.

If you want the old behavior back, you can opt into an idle timeout via /config. But the new default treats your input as required rather than optional. For unattended or scripted sessions where auto-continuation made sense, you’ll need to turn it back on explicitly.

Background agent fixes

The bulk of this release is reliability work for background agents and sessions. Key fixes:

Background sessions no longer silently stop mid-turn after sleep/wake or when reopening a stalled session. Previously, a laptop going to sleep during an agent run could leave the session in a hung state on wake.
A stale daemon.lock file left by a crash no longer prevents background agents from starting again.
Subagents cut off by a rate limit before producing any output now fail cleanly instead of returning an empty result.
Background agents can no longer be taken over by older builds; handover now uses embedded build timestamps to determine which version is newer.
Background-agent daemon respawn no longer re-runs a turn that was cancelled with Esc before the stall happened.
Control bytes from background-agent output no longer reach the terminal directly.

There are also fixes for plugin loading from git worktrees, the /mcp server list not tracking focus correctly for screen readers, and voice dictation showing a misleading “Voice connection failed” message when recording captures no audio but the connection itself is fine.

Terminal rendering under tmux

Synchronized terminal output is now enabled under tmux 3.4+, which eliminates the rendering flicker that tmux users were seeing. This is a tmux-specific fix; if you’re not running inside tmux you won’t notice any difference.

Version 2.1.201 (same day)

A companion release the same day fixed one thing: Claude Sonnet 5 sessions no longer use the mid-conversation system role for harness reminders. This is an internal change to how the harness communicates with the model; no user-visible behavior should change.

Full changelog: code.claude.com/docs/en/changelog