clauding.de
DE
• 2 min read AI-generated

Claude Code 2.1.139: Agent View Puts All Your Sessions in One Place

Copy article as Markdown

The biggest feature update in weeks: Agent View, the new /goal command, and over 40 bug fixes make Claude Code feel noticeably more mature.

Claude Code Release Agent View Anthropic
Featured image for "Claude Code 2.1.139: Agent View Puts All Your Sessions in One Place"

Anthropic shipped Claude Code 2.1.139 on Sunday evening — and this one’s a big deal. Beyond 40+ bug fixes, it introduces two features that could meaningfully change how you work with Claude Code day to day.

Agent View: All Sessions at a Glance

The headline feature is Agent View (Research Preview). Run claude agents and you get a single list of every Claude Code session — running, waiting on you, or done. If you regularly spin up multiple agents across worktrees or background tasks, you know the pain of losing track. Agent View solves that cleanly.

/goal: Claude Works Until It’s Done

The new /goal command is equally exciting. You set a completion condition and Claude keeps working across turns until it’s met. A live overlay panel shows elapsed time, turn count, and token usage as it goes. It works in interactive mode, with -p, and in Remote Control.

What Else Is in There

The changelog is long. Some highlights worth noting:

  • Hook improvements: New args field for hooks (exec form, no shell), continueOnBlock option for PostToolUse hooks
  • MCP improvements: Stdio servers now receive CLAUDE_PROJECT_DIR in their environment, /mcp Reconnect picks up .mcp.json edits without a restart
  • Plugin details: claude plugin details <name> now shows component inventory and estimated per-session token cost
  • Compaction: The compaction prompt now explicitly asks the model to preserve sensitive user instructions
  • VS Code: Press Cmd/Ctrl+Shift+T to reopen the most recently closed session tab

On the bug fix side, a particularly nasty deadlock was fixed where expired credentials could completely block claude auth login with no way to recover. Unbounded memory growth from broken SSE streams on MCP servers was also addressed — response bodies are now capped at 16 MB per frame.

My Take

Agent View and /goal complement each other perfectly: set a goal, let Claude work, and keep track of everything through Agent View. It’s a tangible step toward more autonomous workflows in Claude Code — without losing control.

The sheer volume of bug fixes also shows Anthropic is listening closely to community feedback. From CJK rendering to Windows Terminal quirks to plugin symlinks — they’ve been busy polishing the edges.

Sources: