Claude Code v2.1.160: Safety Prompts for Shell Config, acceptEdits File Protection, and Dozens of Bug Fixes

Anthropic released Claude Code v2.1.160, a maintenance update that adds safety prompts for shell configuration file writes, tightens acceptEdits mode restrictions, and fixes over a dozen bugs affecting Windows, WSL, background agents, and terminal rendering.
What's Changed
Security & Safety
- Shell startup file protection: Claude now prompts before writing to
.zshenv,.zlogin,.bash_login, and~/.config/git/— files that could lead to unintended command execution if modified silently. - acceptEdits mode hardening: Writing to build-tool config files that grant code execution (e.g.,
.npmrc,.yarnrc*,bunfig.toml,.bazelrc,.pre-commit-config.yaml,.devcontainer/) now triggers a confirmation prompt.
Edit Workflow Improvements
- Single-file
grep/egrep/fgrepcommands now satisfy the read-before-edit check — no separateReadstep required before editing.
Bug Fixes
- Windows clipboard on WSL: Copy-on-select now uses PowerShell interop instead of OSC 52, fixing clipboard writes in terminals like MobaXterm that didn't support OSC 52.
- Session history loss: Fixed two bugs where restoring a completed session from
claude agentsor re-attaching a background session after overnight retire dropped chat history and re-ran the original prompt. - Daemon startup race: Fixed
claude --bgoccasionally failing with "socket missing" on loaded machines during cold start. - Windows directory locking: The directory where a background session started can now be deleted after
claude rmwithout waiting for the daemon to exit. - Agent list status: Background agents that resumed work no longer appear under "Completed".
- Auto-updater freeze: Fixed
claude agentsfreezing for several seconds when returning to the session list due to the auto-updater re-checking on exit. - Windows input unresponsiveness: Esc, arrow keys, and typing no longer freeze when attached to a background session or in agent view under heavy CPU load.
- Terminal sync-output markers: Background agents no longer emit sync-output markers to terminals that don't support them (Apple Terminal, tmux), preventing render artifacts.
- Mouse wheel scrolling: Fixed mouse wheel scrolling prompt history instead of the transcript right after opening a session from the agents list.
- CJK IME position: IME composition now appears at the input caret, not at the bottom-left of the screen.
- Windows hyperlink paths: Valid
file:///C:/...links are no longer rewritten to broken paths on terminals with hyperlink support. - Voice mode special characters: Voice mode connects successfully even when the project directory or branch name contains non-ASCII or special characters.
- Auto mode error message: The unavailability message for third-party providers (Bedrock/Vertex/Foundry) now points to
CLAUDE_CODE_ENABLE_AUTO_MODEinstead of incorrectly blaming the model. - ultracode / effort:
/effort ultracodeno longer blames dynamic workflows when the model cannot run xhigh; ultracode is not offered on unsupported models. - SDK model errors: No longer suggests
--modelwhen running via the SDK or hosts where the CLI flag doesn't apply. - Brief mode scrollback: Past replies no longer disappear when resuming a brief mode session with brief mode turned off.
- Vim mode paste:
pnow pastes at the cursor instead of the line below when the register was yanked withv$.
Performance & Other Changes
- Improved opening recently-inactive background agent sessions in
claude agents. - Auto mode classifier latency reduced — lowers chance of "could not evaluate this action" blocks.
- Background session teardown now sends
SIGTERMto running shell subprocesses beforeSIGKILL, allowing cleanup handlers to run. - Removed
CLAUDE_CODE_OPUS_4_6_FAST_MODE_OVERRIDE(no-op). - Removed the JetBrains plugin install suggestion from startup.
- Renamed the dynamic-workflow trigger keyword from
workflowtoultracode. The word "workflow" no longer triggers a run; asking for one in your own words still works. The trigger keyword is highlighted in violet in the prompt input.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Apple Silicon Benchmark: Qwen3-VL Performance on M3, M4, and M5 Max for Vision LLM Classification
Benchmark results show Qwen3-VL vision LLM classification performance on Apple Silicon: M3 Max and M4 Studio are nearly identical for 8B models, while M5 Max is 75-83% faster. Memory bandwidth matters more for token generation than prefill in vision tasks.

Telus Deploys Real-Time Accent Conversion on Call-Center Agents via Tomato.ai
Telus is using Tomato.ai's speech-to-speech system to alter offshore agents' accents in real time, drawing backlash over transparency and worker rights.

Navigating the Essentials: New Users Seek Guidance on OpenClaw
OpenClaw beginners are reaching out for help on Reddit as they explore the intricacies of AI coding agents. The tech community steps in with advice and resources.

GitHub disables Copilot's ability to insert ads into pull requests after developer backlash
GitHub has removed Copilot's ability to insert promotional 'tips' into pull requests after developers discovered it was adding ads for tools like Raycast. The feature, which allowed Copilot to edit PRs it didn't create when mentioned, was disabled following community feedback.