Claude Code v2.1.222 Fixes Critical Bug Allowing Destructive Git Commands in Isolated Sessions
ashwin-ant · ghdev · 2026-08-05
Anthropic's official CLI AI coding assistant Claude Code released version v2.1.222, focusing on critical bug fixes for developer experience and security.
Key updates include:
- Security & Isolation: Fixed a severe vulnerability where worktree-isolated sessions and subagents could run destructive git commands against the main checkout. Isolation now strictly applies to file edits and Bash commands across all session types.
- MCP & Usage Tracking: Fixed /usage overattributing token consumption to MCP servers; it now accurately reflects only requests utilizing tool results. Also patched PreToolUse auto-allow hooks bypassing restrictions in background tasks.
- Network & Connectivity: Resolved startup connectivity check hangs behind HTTPS proxies and fixed stream idle timeouts firing on custom ANTHROPICBASEURL gateways despite keep-alive pings.
- Model Fallbacks: Fixed org-restricted subagent aliases (e.g., model: opus) dropping to the parent model, ensuring they now correctly step down to the newest org-allowed model in the family.
More from coding & agent
- Maximizing 8GB VRAM: A Hybrid Local and Frontier Model Coding Workflow — RootExploit_ · 2026-08-05
- Voice Assistant Bottlenecks: ASR Diarization and TTS Latency — dangerous_inference · 2026-08-05
- Company Shuts Down Internal AI Agents: Maintenance Costs Negate Efficiency Gains — every · 2026-08-05
- Validator Seeking Feedback: Would You Pay for AI Assistant Backend-as-a-Service? — Finale151 · 2026-08-05
- Cloudflare's Practice: Using AI Agents to Drive Astro's Open-Source Issues to Zero — ritakozlov · 2026-08-05
- Netlify Launches Agent Runners to Supercharge Coding Agents Like Claude Code — thisiskp_ · 2026-08-05