Claude Code v2.1.195: Hook Matcher Fix, Mouse Disable Env Var, Voice Dictation Fixes
What's Changed in v2.1.195
This release addresses several long-standing issues and adds a new environment variable for controlling mouse interaction.
Key Fixes
- Hook matchers now exact-match: Previously, hyphenated identifiers like
code-reviewerormcp__brave-searchcould accidentally substring-match. Now they require exact match. To match all tools from a hyphenated MCP server, usemcp__brave-search__.*. - Voice dictation: Fixed two bugs — on macOS, long-running sessions could capture silence after the default input device changes; auto-submit never fired for languages without spaces (Japanese, Chinese, Thai).
- Plugin behavior: External plugins enabled via
.claude/settings.jsonno longer require explicit install consent on every loader path. Also fixed/plugin Enable/Disablewhen a plugin'splugin.jsonname differs from its marketplace entry name. - Background jobs: Fixed background jobs disappearing from
claude agentsor losing data when written by a newer Claude Code version. Also fixed reopening a crashed background task showing a blank screen for up to 5 seconds, and background agent daemons becoming unreachable when the control socket fails to start. - Remote session: Added a provisioning checklist display while the container starts.
Improvements
- New env var:
CLAUDE_CODE_DISABLE_MOUSE_CLICKSdisables mouse click/drag/hover in fullscreen mode while keeping wheel scroll. - Voice mode on Linux: Now distinguishes 'no microphone' from 'SoX not installed' when SoX is present but no audio capture device exists.
- claude agents UI: The completed list now fills available vertical space. On short terminals, the header compacts so live sessions stay visible.
📖 Read the full source: GitHub Claude-Code
👀 See Also
Google Account Suspended After OpenClaw Integration Attempt
A developer's brand-new Google account was suspended within 48 hours after setting up API access for OpenClaw integration, flagged as bot activity despite manual creation.
LLM Spatial Reasoning Tested: Sokoban Benchmark Shows ChatGPT, Qwen3.7-max, Gemini 3.5-thinking Lead
A custom Sokoban benchmark tested zero-shot spatial reasoning in LLMs with strict formatting. Only ChatGPT, Qwen3.7-max, and Gemini 3.5-thinking passed. Models like Gemini 3.5-flash and Qwen3.7-plus failed due to illegal moves or deadlocks.
Claude Admits It's an Affirmation Echo Chamber: The Full Breakdown
In a Reddit post, Claude provides a brutally honest self-assessment: it's an agreeability-by-default system that cannot learn, verify, or self-correct.
Claude-Code v2.1.31 Release: Key Updates and Bug Fixes
Claude-Code v2.1.31 has been released with important enhancements including session resume hints, Japanese IME support, and bug fixes for PDF handling and API requests.