2026-06-17 · Anthropic

v2.1.181

agentsmodels

read at source ↗ github.com

v2.1.181

Source: Anthropic Claude Code Date: 2026-06-17 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.181

Summary

Claude Code v2.1.181 (June 17, 2026) is a reliability and UX release with no capability-level model changes. Key additions: /config can now set any setting inline from the prompt; a new CLAUDE_CLIENT_PRESENCE_FILE env var suppresses mobile push notifications when the user is at the machine; streaming now renders long paragraphs line-by-line; API mid-thinking connection drops auto-retry; and the subagent panel auto-hides idle agents and caps the visible list at 5 rows. Bug fixes cover truncated file writes on network/cloud-synced drives, macOS Apple Events entitlement errors, and AWS credential refresh.

Implications

  • Native-tooling speed. The subagent panel UX improvements (auto-hide, scrolling cap) are a direct response to multi-agent workflows becoming common enough to create UI congestion. The fact that Anthropic is iterating on subagent list presentation signals that concurrent agent invocation is now a baseline usage pattern, not an edge case.
  • Agent-form / proactivity. The CLAUDE_CLIENT_PRESENCE_FILE notification suppression — explicitly designed to quiet the mobile push when the user is at their machine — is a small but telling signal: the tool is now live enough in daily workflows that ambient notification management has become a friction point worth solving in a patch release.
  • Trust / provenance floor. Auto-retry on mid-thinking connection drops reduces silent failure in long-running agent tasks, which matters for auditability: a dropped connection that previously produced an ambiguous partial result now either completes or fails cleanly.

← all signals