Claude Code 2.1.209 to 2.1.216: /fork and /verify Change

Two updates can break existing setups: /fork now starts background sessions instead of subagents, and /verify no longer runs automatically. On top of that come a new sandbox setting, a performance fix for long sessions, and protective limits against uncontrolled subagents, all in versions 2.1.209 to 2.1.216. After reading this article, you will know whether you need to adjust anything.

A note on the sources: the official changelog does not give details on the security fixes in 2.1.214, exact release intervals, or the technical mechanics of EndConversation. Wherever this article only has the rough statement from the changelog available, that is marked with the short tag "(per the changelog)."

What is changing with /fork and background sessions?

As of version 2.1.212, the /fork command no longer starts in-session subagents but instead creates a full copy of the conversation as a standalone background session. This copy keeps running in parallel to the main session, independent of its context and resources.

Also new in this release series: MCP tool calls that run longer than two minutes automatically move to the background.

What this means for you: you can use /fork to start parallel experiments without blocking the main session or sharing its context.

Why are there now limits for subagents and WebSearch calls?

As of version 2.1.212, Claude Code limits WebSearch calls and the creation of new subagents to 200 each per session by default. The changelog cites protection against "runaway delegation" as the reason: uncontrolled chains of ever more subagents or search queries that drive up time and cost.

In addition, there is a new command, claude auto-mode reset, which resets the state of auto mode. What this means for you: if you run Claude Code in automated pipelines with little oversight, you should check whether the new default limits are enough for your own workflow.

What does the new sandbox setting in version 2.1.216 mean?

Version 2.1.216 introduces the setting sandbox.filesystem.disabled. It lets you use the sandbox's network control without also activating filesystem isolation.

Previously, the two isolation layers were more tightly coupled: anyone who only wanted to restrict network access also had to accept filesystem restrictions. What this means for you: you can now set network rules specifically, while the agent continues to access the local project directory unhindered.

What performance fix does version 2.1.216 bring for long sessions?

Version 2.1.216 fixes a performance drop in long sessions: message normalization previously grew quadratically instead of linearly with conversation length. In practice, this made Claude Code noticeably slower the more messages accumulated in the history.

The same version brings additional stability fixes for background sessions and git worktree management (per the changelog). What this means for you: there is nothing to configure, a simple update to 2.1.216 is enough.

What security fixes were added in version 2.1.214?

Version 2.1.214 closes security gaps in the permission checks for Bash and PowerShell commands (per the changelog). The same version introduces the EndConversation tool, which can actively end a conversation in cases of abuse.

Both changes concern safeguarding against unintentional or malicious use of system commands. What this means for you: if you use Claude Code in production, you should update to at least version 2.1.214.

What is changing with /verify and /code-review?

As of version 2.1.215, /verify and /code-review no longer run automatically, only when explicitly invoked. Previously, both commands could be triggered on their own in certain situations.

What this means for you: if you previously planned /verify or /code-review as an automatic workflow step, you now have to actively build the call into your own scripts or slash commands.

What does this mean for teams using Claude Code in production?

Versions 2.1.209 to 2.1.216 bundle two themes: more control over resource consumption and more stability in long or parallel sessions. The new default limits for WebSearch and subagents reduce the risk of unexpectedly high costs, while the sandbox switch allows more precise network rules.

/fork now starts background sessions, and /verify no longer runs on its own: both changes affect existing workflows. A comparison with other AI coding tools is available in our article Claude Code, Cursor, and Copilot Compared.

Check after the update:

• Does your workflow use /fork as a replacement for in-session subagents? It now runs as a separate background session.
• Is /verify or /code-review part of an automated script? The call now has to be made explicitly.
• Does your pipeline need more than 200 WebSearch or subagent calls per session? Then the new default limits apply.

Anyone who wants to integrate such tools into company processes, not just use them individually, needs sandbox rules and a CI connection. On our page about AI development, we show how we build and secure such setups for clients.

Version	Date/series	Key change
2.1.209	July 2026 update series	Bug fix: dialogs (e.g., model selection) were incorrectly blocked in background sessions
2.1.210	July 2026 update series	Live counter for long-running tool calls; isolation fixes for worktree subagents
2.1.212	July 2026 update series	/fork starts background sessions instead of in-session subagents; new limits for WebSearch (200) and subagent spawns (200); command claude auto-mode reset
2.1.214	July 2026 update series	Security fixes in permission checks (Bash, PowerShell); new EndConversation tool
2.1.215	July 2026 update series	/verify and /code-review only run on explicit invocation, no longer automatically
2.1.216	July 20, 2026	Sandbox setting sandbox.filesystem.disabled; performance fix against slowdowns in long sessions; stability fixes for background sessions and git worktrees

Frequently asked questions about the Claude Code update

How often do Claude Code updates come out?

Between 2.1.209 and 2.1.216 alone there are eight numbered releases, a very short release cadence of several updates per week. The changelog does not consistently give exact intervals in days (see the transparency note above).

Do I have to update Claude Code manually?

The official changelog makes no blanket statement on this. In practice, the CLI usually checks automatically for new versions on startup. If you use fixed version numbers in CI pipelines, you should still double-check your own update mechanism.

What changed since version 2.1.208?

Our previous analysis of version 2.1.208 describes the state before that. The difference in short: 2.1.209 to 2.1.216 bring the new limits, the sandbox setting, and the changes to /fork and /verify.

What is the community tracker claude-code-changelog?

The repository github.com/marckrenn/claude-code-changelog is an unofficial, community-maintained project. According to its own description, it documents all 461 releases to date and additionally extracts system prompts and feature flags for individual versions, but it is not an official Anthropic source.

Do I have to update immediately because of the security fixes?

The changelog classifies the changes in 2.1.214 as security fixes for Bash and PowerShell permission checks, without quantifying the risk in numbers. Without a precise risk assessment, there is little reason not to update to at least 2.1.214 promptly.

Does the EndConversation tool change anything about my normal usage?

According to the changelog, EndConversation is meant for cases of abuse where a conversation has to be actively ended. For regular use, this should barely be noticeable; the changelog does not disclose technical internals (see the transparency note above).

Software keeps evolving continuously. All information in this article reflects the state as of July 21, 2026, and may change with future Claude Code versions.

Sources

• Official Claude Code changelog (code.claude.com)
• CHANGELOG.md in the GitHub repository anthropics/claude-code
• Community tracker claude-code-changelog (unofficial)