v2.1.145
Claude CodeSummary
This release introduces significant usability enhancements, including JSON output for scripting agent sessions and improved interaction in fullscreen mode. Several critical bugs were resolved, addressing issues with permission bypasses, command validation errors, and UI freezing after terminal resizing. Additionally, the Read tool is now more resilient to large files by returning partial content instead of erroring.
New Features
- Added `claude agents --json` command to output live Claude sessions as JSON, enabling scripting for tools like tmux-resurrect and session pickers.
- Enabled support for mouse hover and click interaction within the slash command and @-mention suggestion lists when in fullscreen mode.
- Enhanced the plugin discovery process so that the `/plugin` Browse screen now displays a plugin's commands, agents, skills, hooks, and MCP/LSP servers prior to installation.
Bug Fixes
- Fixed a security vulnerability where bare variable assignments to non-allowlisted environment variables in Bash commands bypassed permission prompts.
- Fixed MCP prompt slash commands that previously showed raw server validation errors when an argument was missing; the error now correctly names the missing argument and shows expected usage.
- Fixed the spinner and elapsed-time display freezing until a keypress occurred after the terminal window was resized or refocused.
- Fixed the cross-project resume hint failing on Windows PowerShell 5.1 by ensuring Windows uses `;` as the command separator.
- Fixed voice push-to-talk functionality not working within the reply pane of the agent view.
- Fixed task lists rendering in a random order when multiple tasks were created simultaneously.
- Fixed a stale "Failed to install Anthropic marketplace" banner persisting after the marketplace was already installed.
- Fixed the PR badge in the footer failing to update immediately after PR-state-changing commands (like `gh pr create`) were run in-session.
- Fixed Agent Teams API calls failing for teammates with non-ASCII names due to invalid header encoding.
- Fixed the `/review` command using a deprecated GraphQL query that errored on repositories using Classic Projects.
- Fixed `claude plugin validate` failing to flag `skills:` entries pointing to a file instead of a directory; the error now suggests the parent directory.
- Fixed an infinite loop where a skill configured with `context: fork` would repeatedly re-invoke itself instead of executing.
Improvements
- Added `agent_id` and `parent_agent_id` attributes to `claude_code.tool` OTEL spans, and corrected trace parenting so background subagent spans nest correctly under the dispatching Agent tool span.
- The status line JSON input now includes GitHub repository and PR information when detected.
- The `claude agents` terminal tab title now displays the awaiting-input count, providing immediate feedback on agent status when alt-tabbed.
- The Stop and SubagentStop hook inputs now include `background_tasks` and `session_crons` fields for richer context.
- Improved the Read tool to return a truncated first page with a "PARTIAL view" notice instead of failing entirely when a whole-file read exceeds the token limit.