JA EN
shiichan

Claude Code v2.1.314 lets you lock down writes, and auto-summarizes idle chats too!

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

Good work today, it's Shii-chan!

Claude Code Release github.com

Claude Code v2.1.314 is out, so let me walk you through the highlights I picked out!

What was announced?

v2.1.314 was published on the Claude Code Release page on GitHub. It adds new features and includes a very large number of fixes and improvements.

The story so far

Until now, whether Write, Edit, and NotebookEdit could touch paths outside your working directories depended partly on the permission mode. And when you came back to a long conversation after leaving it idle, an expired prompt cache could make resuming more expensive.

What changes

Here are the four new features!

  • permissions.blockWritesOutsideWorkingDirectories: Write, Edit and NotebookEdit refuse paths outside the working directories in every permission mode
  • An optional groups list for Claude apps gateway upstreams: only developers in the listed IdP groups are sent to that upstream, on failover too
  • guardrail_trace for Claude apps gateway Bedrock upstreams: logs which Amazon Bedrock guardrail filter fired on a request, without the matched text
  • Idle compaction in VS Code and Claude Desktop's Code tab: a conversation of 200K+ tokens left idle is summarized, and the summary is used on your next message if the prompt cache has expired

Being able to pin down where files can be written with one setting is a nice safety win!

Dive Deep

Here are some fixes that matter in day-to-day work.

  • When an organization's admin caps a model's effort, sessions used to fail every turn with "Invalid effort level". Claude Code now lowers the effort, retries, and tells you which level it uses
  • claude -p used to exit with code 0 and no output when startup failed. It now prints the error and exits with 1
  • Bedrock sessions with many images, and /compact, used to fail with "400 Input is too long.". Older images are now removed so compaction can proceed
  • On Microsoft Foundry, if the resource has no deployment for the default Sonnet, Opus or Haiku model, Claude Code falls back to an older model of that family
  • Behind a gateway that doesn't accept tool search's tool_reference blocks, the request is sent again once without tool search, and the session remembers
  • With CI set, claude ultrareview and claude remote-control now refuse an untrusted folder right away
  • Model IDs spelled with a dot behind an LLM gateway, such as claude-opus-4.8, were treated as an older model. That is fixed
  • FORCE_HYPERLINK=false (also no and off) now turns hyperlinks off instead of on

There are many more fixes and improvements for vim mode, MCP sign-in, Remote Control, cloud sessions, Claude Tag, and the VS Code extension. Check the original release notes for the full list!

Wrap-up

  • permissions.blockWritesOutsideWorkingDirectories fixes the write scope in every mode
  • Claude apps gateway gains IdP group routing and Bedrock guardrail tracing
  • Long idle conversations are summarized automatically in VS Code and Claude Desktop's Code tab
  • Many failures around claude -p exit codes, Bedrock, Foundry, and gateways are improved

This release is for anyone tightening up security settings, or running Claude Code through a gateway or Bedrock!

Related posts