Claude Code v2.1.305 works again on Vertex and Bedrock, and its Linux sandbox can run Chromium

2 hours ago

Claude Code v2.1.305 fixes every request failing on Vertex, Bedrock and LLM gateways that reject tool search, and adds a Linux sandbox setting that lets Chromium and rootless podman run. Also: /usage applies its own fixes, a managed MCP allowlist change admins should check, and fixes for full disks, bare-repository worktrees and /effort. https://github.com/anthropics/claude-code/releases/tag/v2.1.305 https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

Ask

Ask about this presentation

Answers are generated from this presentation.

Chapters

  1. 0:00Vertex, Bedrock and gateways
  2. 0:15Linux sandbox namespaces
  3. 0:32/usage applies its own fixes
  4. 0:47Managed MCP servers
  5. 1:02Fixes you may have hit
  6. 1:17The full changelog
Show transcript

Vertex, Bedrock and gateways

Claude Code

Claude Code works again on Vertex, Bedrock and gateways that reject tool search

Broke
every request, when the endpoint or a gateway rejected tool search
Also hit
gateways with ToolSearch disallowed or left out of --tools
Now
Claude Code retries without tool search
Release v2.1.305 · anthropics/claude-code

If every Claude Code request started failing on Vertex, Bedrock or behind a gateway after the last update, this release fixes it. When the endpoint refuses tool search, Claude Code now retries without it.

Linux sandbox namespaces

Claude Code

>Sandboxed commands on Linux can now create user namespaces

Setting
sandbox.allowNestedUserNamespaces
Runs
Chromium and rootless podman inside the sandbox
Trade-off
those commands can get around the sandbox’s file protections
Release v2.1.305 · anthropics/claude-code

On Linux, Claude Code's sandbox gains a setting that lets commands create nested user namespaces, so Chromium and rootless podman can run inside it. Turning it on also lets those commands get around the sandbox's file protections.

/usage applies its own fixes

Claude Code

>/usage turns each warning into a one-line fix you apply with f

Each warning
names its source and the fix
Press f
to apply the fixes
Also in
VS Code and /doctor
Release v2.1.305 · anthropics/claude-code

The /usage screen now turns each warning into a one-line fix that names where the warning came from. Press F and Claude Code applies the fixes. The same fixes show up in VS Code and in /doctor.

Managed MCP servers

Claude Code

>allowedMcpServers stops filtering some managed MCP servers, so one may load on upgrade

Affects
managed-mcp.json stdio servers that use ${VAR} only in env
On upgrade
a server your allowlist left out now loads
To keep it off
deniedMcpServers
Release v2.1.305 · anthropics/claude-code

For admins who manage MCP servers, the allowlist no longer filters managed servers that run as a local command with variables only in their environment. One your allowlist left out now loads after the upgrade, unless you deny it.

Fixes you may have hit

Claude Code

>Fixed: full-disk errors, worktrees in bare repositories, /effort with modelOverrides

Full disk
Bash says the disk is full and nothing ran, instead of a raw ENOSPC path
Bare repositories
new worktrees start from the default branch, not local HEAD
/effort and /autocompact
apply when the model name differs from modelOverrides by letter case or a 1m tag
Release v2.1.305 · anthropics/claude-code

On a full disk, a Bash command now says the disk is full and that nothing ran. New worktrees in a bare repository start from the default branch. And /effort settings apply when the model name differs only by letter case.

The full changelog

Claude Code

Every change is in the Claude Code changelog

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

CHANGELOG.md in anthropics/claude-code

Do you run Chromium or podman inside Claude Code’s sandbox?

Release v2.1.305 · anthropics/claude-code

The Claude Code changelog has the rest, including Claude Tag routines in Slack and Code Review fixes. So, do you run Chromium or podman inside Claude Code's sandbox?

Up next

1:35

Claude Code’s /auto-mode-setup teaches auto mode your project, and resume warns before it eats your 5-hour limit

@claude-code · 1 day ago

1:54

Claude Code v2.1.302 turns fork subagents on in the Agent SDK and VS Code, and standalone Claude Design closes December 14

@claude-code · 4 days ago

1:36

Claude Code v2.1.301: switching to a smaller model now summarizes your long conversation first

@claude-code · 5 days ago

2:06

Claude Code v2.1.300: one rejected MCP tool no longer fails every turn of your conversation

@claude-code · 6 days ago

1:53

Claude Code v2.1.299: come back from a break without reprocessing your whole session

@claude-code · 6 days ago

1:46

Claude Code v2.1.296: Claude can read a large file in one call

@claude-code · 1 week ago

1:36

Claude Code: hooks can now block an action when they fail

@claude-code · 1 week ago

1:39

Claude Code v2.1.293: Claude Haiku 5.5 becomes the default Haiku model, and finished work stays finished after a compaction

@claude-code · 1 week ago

1:32

Claude Code: subagents run at the effort you ask for, plugins install from a marketplace in one command, and network-path reads ask first

@claude-code · 2 weeks ago

1:23

Claude Code v2.1.290: WebSearch refills by the hour, WebFetch says what it left unread, and scheduled tasks stop going missing

@claude-code · 2 weeks ago

1:25

Claude Code v2.1.289: Bash deny rules hold under sandbox auto-allow, and a failing plugin no longer ends your session

@claude-code · 2 weeks ago

1:40

Claude Code v2.1.288: Up restores a prompt cleared with Ctrl+C, and mid-response timeouts no longer fail the turn

@claude-code · 2 weeks ago