Anthropic · 2026-05-11 · notable
Claude Code 2.1.139 — Agent View Lists Every Session in One Place; /goal Locks the Agent on a Condition Across Turns
Anthropic ships a research-preview Agent View that surfaces every Claude Code session (running, blocked, or done) in one list, plus a /goal command that keeps the agent working across turns until a stated condition is satisfied.
Claude Code's May 11 patch adds an Agent View for multi-session monitoring and a /goal command that holds the loop until a condition is met.
What is it?
Version 2.1.139 of Anthropic's Claude Code CLI. Headline additions are an Agent View research preview (a single command, `claude agents`, that lists every running, blocked, or done session) and a `/goal` command that fixes a completion condition the agent must reach before stopping.
How does it work?
Agent View aggregates sessions from interactive, `-p`, and Remote Control modes onto one screen with status. `/goal` sets a target the agent re-evaluates after every turn, with an overlay that tracks elapsed time, turn count, and token usage. The release also adds `args: string[]` exec-form hooks, `continueOnBlock` for PostToolUse hooks, `CLAUDE_PROJECT_DIR` exposure to stdio MCP servers, hot-reload of `.mcp.json` via `/mcp Reconnect`, and `x-claude-code-agent-id` headers on subagent requests.
Why does it matter?
Long-running multi-agent work is where Claude Code users get lost — sessions stall, you forget which terminal owns which task, and the agent quits when it thinks it is done. Agent View gives you a control room; /goal closes the gap between 'be helpful' and 'finish this thing.' Together they nudge Claude Code further toward unattended operation.
Who is it for?
Claude Code power users and multi-agent operators
Try it
npm install -g @anthropic-ai/claude-code@2.1.139 ; claude agents