AI Briefing
KOSign in

Claude Code 2.1.319 adds denyInsteadOfPrompting and background MCP tasks

·2026.11.08 19:35

Key point

The update introduces a setting to deny permission prompts for continuous execution and moves MCP tasks to the background, alongside security fixes for Bash and sandbox escapes.

Details

  • New permissions.denyInsteadOfPrompting setting allows tool calls to be denied instead of waiting for user input, enabling uninterrupted long-running sessions in terminals, subagents, and teammates.
  • MCP tasks in the terminal now automatically move to the background, returning results to Claude upon completion; users can disable this behavior via CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND or CLAUDE_CODE_DISABLE_MCP_TASKS=1.
  • Security patches address a Bash permission bypass using crafted environment prefixes and a sandbox escape vulnerability triggered during worktree creation.
  • /loop behavior changes so self-paced loops end when a turn finishes without rescheduling, with CLAUDE_CODE_LOOP_KEEPALIVE=1 available to restore previous functionality.
  • Auto mode now denies or asks about server-unverifiable calls when telemetry or feature flags are disabled, rather than falling back to the built-in classifier; CLAUDE_CODE_AUTO_MODE_SERVER=0 retains the classifier.
  • New /claude-api latency-optimize command helps identify performance bottlenecks in Claude API code, and /tasks displays a row for finished subagents earlier in the session.
anthropicclaude-codemcpdeveloper-toolssecurity-fixversion-update
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditClaude Code ignores AGENTS.md when CLAUDE.md exists unless explicitly importedRedditDeveloper reports Claude Code plugin submissions with specific features may require verified organization statusGeekNewsAnthropic releases cc-plugin-privacy to block Claude Code from sending user emails in User-Agent headersAnthropic ResearchAnthropic analysis of 207,049 sessions finds Claude usage tracks AI capability, with data analysis and writing dominating scientific researchClaude BlogAnthropic adds security, dev, and creative agents to Claude MarketplaceHacker NewsGoldman Sachs Restricts Claude Code and GitHub to Engineers