2026-06-12 · Anthropic

v2.1.176

securityagentsresearch

read at source ↗ github.com

v2.1.176

Source: Anthropic Claude Code Date: 2026-06-12 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.176

Summary

v2.1.176 is the largest of this trio and spans session continuity, Remote Control hardening, background-agent ergonomics, and several platform-specific fixes. Session titles now generate in the conversation language (pinnable via language setting). A new footerLinksRegexes setting adds regex-matched link badges to the footer. Remote Control sees multiple fixes — silent model switches when connecting from web/mobile, bare numeric disconnect codes, duplicate transcript lines, and sessions not clearing on account switch. Background session reliability gets attention across Windows network paths, corrupted resume IDs, Windows daemon startup with ReadOnly attributes, and cloud sessions going stale after idle.

Implications

  • Fleet-correctness. The background-session cluster of fixes (Windows network paths in persisted state, malformed resume IDs from corrupted state files, cloud sessions failing after idle, the Windows daemon ReadOnly attribute block) reads as Anthropic addressing real-world fleet deployment edge cases. These aren’t synthetic bugs; they’re the tail of a maturing background-agent runtime being exercised in production.
  • Coding-agent host war. Remote Control — connecting from web or mobile into a running Claude Code session — gets four separate fixes. Hardening this path signals that remote/multi-surface session control is a real product surface being actively used, not a prototype feature. The silent model-switch fix in particular (connecting from web could change the active model without notification) is a trust issue as much as a UX one.
  • Trust/provenance hardening. The availableModels enforcement fixes (/fast refusing to toggle to a blocked model, alias picks not redirectable via ANTHROPIC_DEFAULT_*_MODEL) are the companion to v2.1.175’s enforceAvailableModels — the two releases together close both the managed-default escape hatch and the env-var override path. Model governance is now tighter end-to-end.
  • Watch: the hook if-condition fix (documented glob patterns for Read/Edit/Write paths now match correctly) is small but removes a correctness footgun that would have silently misfired policy hooks in production deployments.

← all signals