Claude Code 2.1.161: Parallel Tool Call Independence, Agent Progress Display, and Reliability Fixes
Claude CodeClaude Code version 2.1.161 delivers a meaningful reliability improvement to parallel tool execution: a failed Bash command in a batch no longer cancels the other calls running alongside it. The release also improves visibility into multi-agent work with done/total progress counters in the claude agents view, and brings Linux clipboard improvements using native system tools. A broad set of bug fixes addresses background session reliability, rendering stability, and Windows hook execution.
Sources & Mentions
3 external resources covering this update
Parallel Tool Call Independence
In prior releases, when Claude Code executed multiple Bash commands in parallel, a single failure could abort the entire batch. Version 2.1.161 resolves this: each tool call in a parallel batch now returns its own result independently, even when one command fails. This makes parallel agentic work more resilient: a flaky test command or a file-not-found error in one branch no longer disrupts unrelated operations running at the same time.
Agent Progress Visibility
The claude agents view now displays a done/total counter before work details when tasks are distributed across agents. The peek view surfaces the longest-running item, giving users an at-a-glance understanding of where time is being spent across a parallel workload. For developers running large multi-agent operations, this removes the need to mentally track completion state across many concurrent subagents.
Linux Clipboard Support
Fullscreen mode clipboard now integrates with native Linux clipboard tools: wl-copy (Wayland), xclip, and xsel, when they are available on the system. This closes a gap for Linux users who previously had unreliable clipboard behavior in fullscreen sessions.
MCP and Connector Cleanup
The /mcp panel now collapses claude.ai connectors the user has never authenticated with behind a "Show unused connectors" row, reducing clutter for users who have many connectors configured but only use a subset.
Bug Fixes
This release resolves a substantial backlog of issues:
- Background subagents running in isolation: "worktree" mode were blocked from editing files in background sessions, now fixed.
- Background sessions were booting with a stale model from the daemon environment instead of reading from settings.json.
- Completed subagents could remain stuck showing as "running" when finalization errors occurred.
- EADDRINUSE errors from tools binding Unix sockets under deep CLAUDE_CODE_TMPDIR paths are resolved.
- Windows hooks invoking bash explicitly were failing with "command not found" or "cannot execute binary file."
- OpenTelemetry log events dropped when emitted before telemetry initialization are now properly captured.
- The /autofix-pr command incorrectly reported "cannot run on default branch" in git worktrees.
- The --resume session picker now correctly shows sessions from the current directory when it is not a git worktree.
- claude mcp list/get/add no longer print secrets; ${VAR} references are no longer expanded in output.