In brief: The new version introduces agent management with a clear session overview and automated goal-based workflows, plus numerous stability and UI fixes.
Claude Code 2.1.139 introduces an Agent View that centrally lists all active and completed Claude Code sessions, as well as a new /goal command that allows the AI to work automatically across multiple turns until a condition is met.
New features: The Agent View (Research Preview) displays all Claude Code sessions in a list — running, blocked, and completed — and serves as a central hub for agent management. The new /goal command sets a completion condition and lets Claude work automatically across multiple turns until it is fulfilled; this command works in interactive mode, with the -p option, and in Remote Control. The /scroll-speed command enables live tuning of mouse wheel scroll speed. New is also a claude plugin details <name> command that displays a plugin’s components and the expected token consumption per session.
Workflow improvements: The transcript view now recognizes keyboard shortcuts (?), allows navigation between user prompts ({/}), and toggles the shortcut guide (v). Hook arguments now support a string[] field (exec form) that launches commands directly without a shell and permits path placeholders without quotation marks. The new continueOnBlock option for PostToolUse hooks feeds rejection reasons back to Claude. MCP stdio servers now receive CLAUDE_PROJECT_DIR in the environment; plugin configurations can reference ${CLAUDE_PROJECT_DIR}. /context displays token estimates per skill with model-specific tokenization and names the providing plugin.
Stability and fixes: The compaction instruction requires the model to preserve sensitive user instructions. /mcp reconnect reads .mcp.json changes without restart and shows HTTP status and URL on errors. Remote MCP servers reconnect on transient failures. API requests from subagents now carry x-claude-code-agent-id and x-claude-code-parent-agent-id headers. Fixed issues include a deadlock on expired credentials, missing shell expansion in autoAllowBashIfSandboxed, terminal corruption via hooks, uncontrolled memory growth on HTTP/SSE MCP servers (16 MB limit per frame), wildcard permission matching, missing hot reload for symlinked settings, plugin details errors, and timer issues after stream cancellation. UI fixes address cursor blinking in dialogs, keyboard shortcuts in transcript view, bash history behavior, and inserting multiple images.
Source: ainews-dev.lumi-systems.io · Published 11 May 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.5.2.