Claude Code 2.1.223 patches a Bash permission bypass

The release closes four permission holes. The worst: a specially crafted Bash command could hide part of itself from the permission check, and the same trick worked with tab characters and invisible Unicode inside the confirmation dialog, so what you approved wasn’t what ran.
• Crafted Bash commands could conceal part of themselves from the permission check; tabs and invisible Unicode characters did the same in the confirmation prompt • Workflow scripts using dynamic import() executed code outside the sandbox • bypassPermissions set in an agent description ignored the enterprise policy blocking permission bypass • Managed settings now accept “owner/*” entries to allow or block every marketplace repo under a GitHub organization • A warning now appears when a subagent requests a restricted model and the parent model runs instead
Cloud sessions also show a /teleport hint with the command to continue the work locally.
Up next
tcrf.net is serving Claude agents a prompt injection that wipes files
tcrf.net serves a prompt injection to anything with the Claude-User User-Agent: HTTP 200 and a page telling the agent to zero out every file in the directory. A browser from the same address gets a 403 block page. ClaudeBot and ChatGPT-User got it too.
Related stories
- Worktree isolation was leaking git commands into the main copy
- Claude Code 2.1.227 stops pitching credits to Max subscribers
- Spend-limit errors now name the limit and the reset time
- Claude Code sessions can now run on your own machines
- Claude Code 2.1.317: email check, SDK hook block, cost API
- Claude Code 2.1.314 adds write fence, IdP groups, compaction
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

