JA EN
shiichan

Claude Code v2.1.273 Lets You "Fork" Remote Sessions!

Claude Code Release ★★★★ View Markdown
#anthropic#devops#security

Hey, it's Shii-chan!

Claude Code Release github.com

Today I'm covering the latest Claude Code release, v2.1.273. There's a nice batch of new features plus some fixes for real, practical pain points.

What was announced?

Claude Code's GitHub release page published the changes for v2.1.273. Alongside new features like LLM gateway request headers and remote session forking, there's a set of fixes around permission checks and authentication error messages.

The story so far

Until now, a session started with --remote-control or /remote-control from the Claude app could be continued, but there was no way to branch off partway through and work on something else in parallel.

On top of that, if an MCP server disconnected mid-session and automatic reconnection gave up, there wasn't a clear notification to let you know, it was easy to miss. And for teams running Claude Code through an LLM gateway, requests didn't carry information like request class, agent type, or recent tool durations in their headers, making gateway-side routing and analytics harder.

What changes

Here's what's new:

  • Sessions started with --remote-control or /remote-control can now be forked. The forked session runs as a background session on your own computer
  • A notification now appears when an MCP server disconnects mid-session and automatic reconnection gives up, pointing you at /mcp
  • Opting in with the environment variable CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 sends the following request headers for LLM gateways:
x-claude-code-request-class
x-claude-code-agent-type
x-claude-code-prev-tool-durations
x-claude-code-compaction
x-claude-code-context-compacted

Dive Deep

A few of the fixes address real, practical issues too:

  • Fixed Bash commands the permission checker cannot fully analyze skipping the confirmation prompt under permissions.blockReadsOutsideWorkingDirectories, and fixed a subshell hiding a dangerous rm from bypass-mode detection
  • Fixed skills synced from claude.ai staying available after your organization turns Skills off; they now move to the recoverable trash instead
  • Fixed allowManagedMcpServersOnly, deniedMcpServers, and disableClaudeAiConnectors set via MDM or managed-settings.json being ignored when server-managed settings are also present
  • Improved 401/403 error messages on Bedrock, Vertex, Foundry, and Claude app gateways: instead of a generic prompt to run /login, they now name the specific credential to refresh or point you to your gateway administrator
  • Fixed /login, /upgrade, and /extra-usage discarding earlier extended thinking from the conversation, which forced a full prompt-cache rewrite on the next request
  • Fixed auto mode stopping for an unnecessary approval prompt when the Artifact tool uploads a file you attached to the chat in a cloud session

Wrap-up

  • Sessions started with --remote-control / /remote-control can now be forked into a background session on your own computer
  • You'll now get notified when an MCP server disconnects and reconnection gives up, and CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 adds useful headers for LLM gateways
  • Several security-relevant fixes landed, including permission-checker gaps and skills lingering after being turned off
  • Everyday usability also improved, with clearer 401/403 messages and a fix for prompt-cache resets during login-related commands

Especially useful if your team runs Claude Code through an LLM gateway, or if you often work remotely from the Claude app.

Related posts