2026-06-16 · Anthropic

v2.1.179

securityagentsmodelscommentary

read at source ↗ github.com

v2.1.179

Source: Anthropic Claude Code Date: 2026-06-16 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.179

Summary

Claude Code v2.1.179 is a stability and regression patch. Key fixes: partial responses are now preserved on mid-stream connection drops (spinner no longer sticks at “running tool”); WSL2 mouse-wheel scrolling regression from v2.1.172 is resolved; a Linux-specific bug where a large sandbox glob made the Bash tool description enormous (rendering sessions unusable) is fixed; subagent transcript view (Ctrl+O) and focus management are repaired; and remote session background tasks no longer appear stuck between turns. Plugin loading performance improved for remote sessions.

Implications

  • Fleet-governance: the sandbox glob bug was a silent usability failure for Linux users running agents with broad directory access — worth noting for teams that set permissive allowRead globs in Linux CI.
  • Agent-layer convergence: mid-stream drop recovery (preserving partial responses) is directly relevant to long-running agentic tasks where connection instability would previously lose all in-progress output.
  • Subagent transcript fixes improve observability for multi-agent workflows, a persistent pain point in fleet-level debugging.

← all signals