Claude Code 2.1.310: Monitor tool, cloud roles, sandbox
Claude Code 2.1.310 adds the Monitor tool on Bedrock and Vertex, retries requests gateways reject, and names AppArmor when the sandbox won't start.
Claude Code 2.1.310 has no single headline. Its entry in Anthropic's changelog is dated October 23, 2026. It adds the Monitor tool on Bedrock, Vertex and Foundry, retries requests that some gateways reject, and closes another Windows hook gap. If you run hooks on Windows or route through a gateway, upgrade. Everyone else can treat it as a routine update.
This brief picks up after the 2.1.309 notes. It covers the items that change what you do, not all fourteen bug fixes.
This site is independent and not affiliated with Anthropic. We read the changelog on 2026-10-25. We did not install or run 2.1.310, so nothing below is a hands-on result.
Monitor tool and gateway fixes
The first "Added" entry is about the Monitor tool:
Added the Monitor tool and plugin monitors on Bedrock, Vertex, and Foundry, and when telemetry or nonessential traffic is disabled; long foreground sleeps are now blocked there too
The entry does not describe what the Monitor tool does, so this brief does not either. What it tells you is where the tool now appears: on those three cloud providers, and in setups where telemetry or nonessential traffic is off.
The gateway fix matters more if you sit behind a proxy:
Fixed requests failing behind some gateways and compatible servers that reject the
effort-2025-11-24beta value or mid-conversation system messages; Claude Code now retries without them
An earlier release had a similar problem, which our 2.1.273-2.1.278 notes describe as every request failing "when ANTHROPIC_BASE_URL points at a proxy or gateway". The new entry does not say which gateways were affected.
One more change concerns MCP servers in scripts:
Changed the error for an MCP tool call that times out in a non-interactive session while the server's elicitation is unanswered: it now tells Claude that no answer came and not to repeat the call
In plain terms, the server was waiting for an answer that never came, and Claude now hears that instead of calling the tool again.
Sandbox, permissions and hooks
On Linux, the sandbox entry is about a startup failure:
Linux: Improved
/sandboxand the sandbox startup warning to name Ubuntu's AppArmor restriction on user namespaces, and how to lift it, when it keeps bubblewrap from starting the sandbox
The entry describes a better message, not a change to how the sandbox works. Run /sandbox on an Ubuntu machine where the sandbox will not start, and it should now name the cause. Anthropic's sandbox page describes the problem: "On Ubuntu 24.04 and later, the default AppArmor policy prevents bubblewrap from creating the user namespaces it needs for isolation." The page also gives the profile that fixes it, so we do not repeat the commands here.
Two prompt changes follow. The first is for auto mode:
Improved auto mode's prompt before the first read outside the working directories: it now says when blocking those reads makes shell commands ask for permission
The second is a fix for file read prompts:
Fixed file read permission prompts not showing where a symbolic link leads
So a prompt for a path that is really a symlink now shows the destination. Read it before you approve. For how allow, ask and deny rules fit together, see our permissions guide.
The Windows hook entry continues a thread from 2.1.309:
Windows: Fixed PreToolUse and PermissionRequest hooks letting a tool call through when Git Bash is missing and the hook's command is only quoted strings, such as a quoted path
A hook that lets a tool call through when it cannot run is failing open. Our guide on making hooks fail closed shows the setting that blocks instead.
The Claude moves worth stealing, by email
Workflows tested on real jobs, sent only when they are worth it. Free.
Fixes that protect your work
Three entries touch long sessions. The first is about deleted files:
Fixed background cleanup deleting some recent files, such as pasted images, when
cleanupPeriodDaysis set to a very large value
Our guide on keeping conversations longer than 30 days suggests "cleanupPeriodDays": 3650. The changelog does not say what counts as "very large", so check whether your value is one.
The second is about compaction after an overflow:
Fixed compaction after a "Prompt is too long" error keeping too much of a conversation with dense text, such as base64 or SVG, so the next request could fail again
A different "prompt is too long" bug was fixed in 2.1.306. The third is a memory improvement:
Improved memory use of non-interactive sessions (
-p, Agent SDK) during compaction when the transcript file is very large
None of the three entries gives a number, so we cannot say how much memory is saved.
Cloud sessions and admin roles
Admins get one addition worth reading:
[Cloud sessions] Added Claude Code cloud sessions and Remote Control permissions to Enterprise custom roles, so Owners can control access to each by role
Two related entries change how that switch behaves. Turning on Cloud sessions or Remote Control in admin settings now asks for confirmation and lists the custom roles that have the permission. The cloud environment dialog also opens network access, setup script and environment variables from rows. Both are interface changes, and the entry shows no screenshots. For settings that live in files, see our managed settings guide.
Upgrade readout
| If you | Do this |
|---|---|
| Use PowerShell or Git Bash hooks on Windows | Retest that a blocking hook still blocks |
| Route through a gateway | Upgrade, then watch for requests that failed before |
| Run the sandbox on Ubuntu | Run /sandbox if it fails to start and read the new message |
Set a very large cleanupPeriodDays | Upgrade, then check recent pasted images are still there |
| Run Enterprise with custom roles | Review who holds the Cloud sessions and Remote Control permissions |
| Anyone else | Routine upgrade |
What this brief does not cover
We did not run any command, setting or tool named above. The entry also lists VS Code fixes, Claude Tag fixes, Code Review admin changes and more cloud environment fixes, which we skipped. One improvement changes how a mod's Image draws in some terminals, and our mods guide covers what mods are. The changelog entry is the full list.
Does Claude Code 2.1.310 fix hooks failing open on Windows?
Partly. The changelog says PreToolUse and PermissionRequest hooks no longer let a tool call through when Git Bash is missing and the hook's command is only quoted strings. Other hook failures are not mentioned, so retest your own hooks.
What does Claude Code 2.1.310 change for gateways?
Claude Code now retries without the effort-2025-11-24 beta value or mid-conversation system messages when a gateway or compatible server rejects them. The changelog does not name the gateways that were affected.
Does 2.1.310 change how the Linux sandbox starts?
The changelog does not describe a change to startup itself. It says /sandbox and the startup warning now name Ubuntu's AppArmor restriction on user namespaces, and how to lift it, when that restriction keeps bubblewrap from starting the sandbox.
Sources (checked 2026-10-25)
- Claude Code changelog: the 2.1.310 entry, dated October 23, 2026.
- Configure the sandboxed Bash tool: what the sandbox is.
