Claude-Code v2.1.94 adds Mantle support, fixes critical bugs

Claude-Code v2.1.94 Release Notes
Claude-Code v2.1.94 adds new features and fixes multiple bugs across the AI coding assistant's core functionality, plugin system, and VS Code integration.
New Features
- Amazon Bedrock with Mantle support: Enable with
CLAUDE_CODE_USE_MANTLE=1environment variable - Default effort level change: Changed from medium to high for API-key, Bedrock/Vertex/Foundry, Team, and Enterprise users. Control with
/effortcommand - Slack MCP improvements: Added compact
Slacked #channelheader with clickable channel link for send-message tool calls - Plugin system enhancements: Added
keep-coding-instructionsfrontmatter field support for plugin output styles - Session title hooks: Added
hookSpecificOutput.sessionTitletoUserPromptSubmithooks for setting session titles - Stable plugin skill names: Skills declared via
"skills": ["./"]now use the skill's frontmatternameinstead of directory basename
Bug Fixes
- Rate limit handling: Fixed agents appearing stuck after 429 rate-limit responses with long Retry-After headers - errors now surface immediately
- macOS login: Fixed Console login silently failing with "Not logged in" when login keychain is locked or password out of sync. Error now surfaces and
claude doctordiagnoses the fix - Plugin system fixes: Fixed plugin skill hooks defined in YAML frontmatter being silently ignored; fixed plugin hooks failing with "No such file or directory" when
CLAUDE_PLUGIN_ROOTnot set; fixed${CLAUDE_PLUGIN_ROOT}resolving to marketplace source directory instead of installed cache for local-marketplace plugins - UI/UX improvements: Fixed scrollback showing same diff repeated and blank pages in long-running sessions; fixed multiline user prompts in transcript indenting wrapped lines under ❯ caret instead of under text; fixed Shift+Space inserting literal word "space" instead of space character in search inputs
- Terminal rendering: Fixed hyperlinks opening two browser tabs when clicked inside tmux running in xterm.js-based terminals (VS Code, Hyper, Tabby); fixed alt-screen rendering bug where content height changes mid-scroll could leave compounding ghost lines
- Environment variable handling: Fixed
FORCE_HYPERLINKenvironment variable being ignored when set viasettings.jsonenv - Accessibility: Fixed native terminal cursor not tracking selected tab in dialogs, enabling screen readers and magnifiers to follow tab navigation
- Bedrock compatibility: Fixed Bedrock invocation of Sonnet 3.5 v2 by using the
us.inference profile ID - Session management: Fixed SDK/print mode not preserving partial assistant response in conversation history when interrupted mid-stream; improved
--resumeto resume sessions from other worktrees of same repo directly instead of printing acdcommand - Text encoding: Fixed CJK and other multibyte text being corrupted with U+FFFD in stream-json input/output when chunk boundaries split UTF-8 sequences
VS Code Specific Fixes
- Reduced cold-open subprocess work on starting a session
- Fixed dropdown menus selecting wrong item when mouse was over list while typing or using arrow keys
- Added warning banner when
settings.jsonfiles fail to parse, alerting users that permission rules aren't being applied
📖 Read the full source: GitHub Claude-Code
👀 See Also

AMD Ryzen AI NPUs Gain Linux LLM Support via Lemonade 10.0 and FastFlowLM
AMD Ryzen AI NPUs now support running large language models on Linux through Lemonade 10.0 server with FastFlowLM runtime, requiring Linux 7.0 kernel or AMDXDNA driver back-ports.

Two new models appear on OpenRouter, possibly DeepSeek V4 variants
Two new models named healer-alpha and hunter-alpha have appeared on OpenRouter, with specifications matching leaked details about DeepSeek V4. Initial testing shows both models perform well in roleplay scenarios with no message filtering and faster token generation than GLM 5.0.

Context Quality Degradation in AI Agents: Hallucination Rates Increase with Token Count
Testing shows hallucination rates increase from ~3% at 10K tokens to ~28% at 200K tokens, with recall accuracy dropping below 90% for early-session information once context exceeds 50K tokens.

Claude Code 2.1.76 adds MCP elicitation, worktree improvements, and fixes for context limits
Claude Code version 2.1.76 introduces MCP elicitation support for structured input during tasks, adds worktree.sparsePaths for large monorepos, and fixes 'Context limit reached' errors on 1M-context sessions. Version 2.1.75 made 1M context windows default for Opus 4.6 on Max, Team, and Enterprise plans.