Claude Code v2.1.309 Patches Sandbox Escapes, Symlink Deny Rules, and MCP Config Tampering

✍️ OpenClawRadar📅 Published: October 23, 2026🔗 Source
Ad

Claude Code v2.1.309 is a security-heavy release that closes several sandbox escape vectors and permission-check bypasses on macOS, Linux, and Windows. The fixes land across Bash sandboxing, MCP config protection, worktree isolation, and hook execution.

Sandbox and filesystem fixes

  • Sandboxed Bash commands on macOS and Linux no longer run the shell's startup files outside the sandbox before each command.
  • macOS sandbox denyWrite glob patterns now apply at a path's real location when the pattern's folder is reached through a symlink.
  • Sandboxed commands and auto-accepted edits can no longer modify the MCP config file passed with --mcp-config.
  • Profile store write protection now applies when the config directory is reached through a symlink.

Permission and isolation fixes

  • Bash permission checks no longer trust a variable set by a command whose redirect can fail to open.
  • Compound commands that change directory before running git now prompt for permission even when the cd is wrapped or prefixed — matching behavior for a plain cd.
  • An Edit(path) deny rule now blocks a Bash redirect to that path when the command also holds a subshell.
  • Worktree-isolated sessions refuse locale-prefixed commands and zsh arithmetic commands that the isolation guard should reject.
  • Split-pane agent teammates now inherit the leader's tool deny and --tools rules.
Ad

Session and MCP fixes

  • A disabled MCP server is no longer reconnected by a later permission prompt in print and SDK sessions.
  • Worktrees and branch checkouts created by Claude Code now ignore git replace refs, consistent with its other git operations.
  • Sessions stop compacting repeatedly after reading files under a very large nested CLAUDE.md; a nested CLAUDE.md over the size limit now loads only its start.
  • Messages from other agents and sessions still waiting to be delivered now appear in full under the transcript (ctrl+o), marked as waiting.

Hooks, skills, and UI

  • PowerShell hooks now block when the hook's program exits with code 2, and Windows PreToolUse and PermissionRequest hooks with "shell": "powershell" no longer fail open when Git for Windows isn't installed.
  • A mod's turn.step hook naming a model by alias (e.g., haiku) now resolves to its model instead of failing every request.
  • Skills in --add-dir folders are now reachable as <folder>:<name> when another skill holds the same name.
  • ctrl+enter (send now) behavior is fixed both before a response starts and while a prompt's hooks are running.
  • Long tool output previews are no longer cut through an emoji, keeping session transcripts parseable by jq.

If you run Claude Code in sandboxed or worktree-isolated modes, upgrade. The symlink and --mcp-config fixes are the ones to prioritize.

📖 Read the full source: GitHub Claude-Code

Ad

👀 See Also