Anthropic: Claude Code v2.1.139 — Agent View showing all sessions + /goal command for autonomous completion

Anthropic has released Claude Code v2.1.139, an update that leans heavily into parallel session visibility and autonomous task completion. The headline feature is Agent View in Research Preview status — launched with the claude agents command, it displays a unified list of all sessions: running, blocked (typically awaiting user approval), and completed. Documentation is available at code.claude.com/docs/en/agent-view.

What does the /goal command bring?

The new /goal command sets a completion condition toward which Claude works across multiple turns without further user intervention. It functions in interactive mode, via the -p flag when invoked from a script, and in the Remote Control interface. During execution an overlay panel is visible with three key metrics: elapsed time, step count, and total tokens consumed — making it easier to track the cost of autonomous runs.

Plugin system and configuration

This release introduces claude plugin details <name>, which shows the component inventory of a given plugin and a projected per-session cost in tokens. MCP stdio servers now receive the environment variable CLAUDE_PROJECT_DIR (as hooks do), and plugin configurations can reference ${CLAUDE_PROJECT_DIR} in commands. Hooks gained an args: string[] field (exec form) that runs commands directly without a shell layer — eliminating the need to quote paths with spaces. A continueOnBlock field for PostToolUse was also added; set to true, it returns the block reason to the model and continues the turn.

Transcript navigation and MCP protection

The transcript viewer now has keyboard shortcuts: ? for the shortcuts panel, { and } to jump between user turns, and v to toggle the auxiliary panel. The /context command now calculates per-skill token estimates using the model’s actual tokenizer and shows the plugin name that a skill comes from.

On the protocol side, HTTP and SSE MCP servers now enforce a 16 MB cap per SSE frame as protection against unbounded memory growth. Sub-agent API requests send x-claude-code-agent-id and x-claude-code-parent-agent-id headers, and OTEL spans carry matching attributes. Remote Control, /schedule, and claude.ai MCP are automatically disabled when ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN is set, preventing mixed authentication contexts.