In a nutshell: The version improves hook integration for terminal events, adds identity federation support, and closes 20+ bugs in permission dialogs, history management, and renderer logic.
Claude Code 2.1.141 extends terminal output with desktop notifications and window titles, introduces new environment variables for GitHub and workspace authentication, and fixes numerous interface and behavior issues.
New features and configuration: The terminalSequence field in hook JSON enables triggering desktop notifications, window titles, and alert sounds without a controlling terminal. The new environment variable CLAUDE_CODE_PLUGIN_PREFER_HTTPS allows cloning GitHub plugin sources over HTTPS instead of SSH — relevant for environments without GitHub SSH keys. ANTHROPIC_WORKSPACE_ID supports workload identity federation and scopes minted tokens to a specific workspace when the federation rule spans multiple. The command claude agents --cwd <path> restricts the session list to a directory. The feedback system can now include current sessions (last 24 hours or 7 days) for issues spanning multiple sessions.
Improvements in usability and visibility: The rewind menu gains a “Compact up to here” option to condense earlier context while retaining newer turns. The auto-mode permission dialog now explains when a permissions.ask rule triggered the prompt. The “View diff in your IDE” feature was restored for file edit permission prompts with a connected IDE. Background agents now start with the current permission instead of falling back to default mode. Agents that complete their work but leave a background shell running are moved to “Completed”. The spinner during long thinking pauses turns yellow after 10 seconds to signal that Claude is actively working.
Interface and navigation: Plugin menu navigation was improved: right/tab key switches tabs, up arrow moves to the tab bar, and tab headers and search field are clickable in fullscreen mode. Multiple bugs were fixed: background page queries now use the main model when no Haiku override is set; Windows daemon errors now show the underlying error message instead of opaque failures; wrapper-launched agents correctly display the dashboard instead of the agent type list; deleted working directories when opening crashed sessions avoid redundant dispatches; custom ANTHROPIC_BASE_URL gateways correctly name background jobs.
Fixed bugs in input and dialogs: The /model command in a session no longer changes the auto-compact threshold in other sessions. Switching permission mode during open tool permission prompts closes them correctly when the new setting allows the tool. Enter during permission/dialog prompts is no longer sent as text input. Hooks receive the correct transcript path after worktree switch. Markdown tables with cell wrapping render again as grids instead of key-value layout. Cancelled prompts are no longer removed from arrow-up history. Ctrl+C interrupts turns in vim INSERT/VISUAL mode. Alternative chat submit keybindings work again. Prompt suggestions no longer disable themselves with configured output style. The spinnerVerbs setting is respected in turn completion messages. Light ANSI theme now correctly uses black instead of invisible white for diff context lines. Error overlays show the original error message instead of minified bundle code.
Source: ainews-dev.lumi-systems.io · Published 14 May 2026
Lumi AI News — AI-assisted curation according to Art. 50 EU AI Act. Paraphrase and classification through Lumi News Pipeline v1.5.2.