v2.1.139

Source: Anthropic Claude Code Date: 2026-05-11 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.139

Summary

Claude Code v2.1.139 (May 11, 2026) is a substantial feature release. The two headline additions are the agent view (claude agents — a unified list of all running, blocked, and completed sessions, Research Preview) and the /goal command (a completion-condition loop that keeps Claude working across turns with a live elapsed/turns/tokens overlay). The release also adds /scroll-speed, improved plugin introspection (claude plugin details <name> with token-cost estimates), transcript navigation shortcuts, a hook args: string[] exec form, continueOnBlock for PostToolUse hooks, and CLAUDE_PROJECT_DIR injection into MCP stdio servers. There are roughly 30 bug fixes spanning scroll behavior, settings hot-reload, MCP reconnection, memory growth, Windows path handling, and emoji/CJK rendering.

Implications

• Agent orchestration thread. claude agents as a unified session view and /goal as an autonomous-loop primitive are the most significant orchestration-layer additions since Remote Control. Together they move the interactive CLI toward a persistent agent runtime — a session is no longer ephemeral but observable and continuable.
• Plugin ecosystem maturity. Token-cost-per-plugin estimates and cleaner hook/MCP introspection in /plugin and /doctor signal Anthropic hardening the plugin economy: operators can now reason about context budget before deploying a plugin composition.
• Hook expressiveness. The args: string[] exec form and continueOnBlock option make hooks composable with structured rejection feedback — the agentic control-flow surface is growing without requiring Remote Control infrastructure.

Watch: Whether claude agents (Research Preview) graduates to stable and whether /goal becomes the default mode for long-running tasks in the managed-agent layer.