v2.1.207

Source: Anthropic Claude Code Date: 2026-07-11 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.207

Summary

Claude Code v2.1.207 opens auto mode to Bedrock, Vertex AI, and Foundry without the prior opt-in env var (opt-out via disableAutoMode), switches those platforms’ default model to Opus 4.8, and ships a long list of stability fixes: terminal freezing on long streamed output, a background-agent-teams crash loop from malformed mailbox messages, worktree config left dangling in .git/config, and a shell-injection fix restricting ${user_config.*} in plugin hook/monitor commands to exec-form only.

Implications

• Continues the autonomy-sprint arc tracked this week: auto mode graduating from opt-in-only to on-by-default across all three enterprise cloud platforms is a capability-availability widening, arriving days after v2.1.205/206 hardened the consent and approval boundaries around exactly this kind of unattended execution.
• The plugin ${user_config.*} shell-injection fix is another instance of the “wall the exposure, smooth the friction” pattern: tightening a concrete injection vector in hook/monitor commands while simultaneously loosening auto-mode’s default-on surface.
• Multiple background-agent-teams and worktree-resume fixes are incremental hardening of the same multi-agent runtime plumbing (mailbox crash loops, blank resume after cold reopen) that’s been the recurring maintenance surface across recent releases.