v2.1.183

Source: Anthropic Claude Code Date: 2026-06-19 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.183

Summary

Claude Code v2.1.183 is a substantial safety and stability release. The headline changes block destructive operations in auto mode — git reset --hard, git clean, git commit --amend on commits not made by the agent, and IaC destroy commands — unless explicitly requested. A batch of agent-layer fixes addresses subagent spawning errors, WebSearch in subagents, tmux teammate reliability, background task premature termination, and webhook/scheduled-task inputs being misclassified as keyboard events.

Implications

• Floor/trust: Blocking destructive git and IaC commands by default is a direct autonomy floor enforcement — the tool now constrains itself rather than relying on prompt-level instructions. This is the pattern worth watching across the tool class.
• Agent layer: The subagent-spawn 400 error fix, WebSearch fix in subagents, and background-task kill fix all indicate the multi-agent execution path is still hardening; reliability at depth is not yet stable.
• Capability clocks: Attribution settings (omitting session URLs from commits) and deprecation warnings for model frontmatter signal maturing operator controls for fleet deployment.