Claude Code 2.1.301-2.1.302: auto mode model, subagents

Claude Code 2.1.301 is dated October 15, 2026 in Anthropic's changelog, and 2.1.302 is dated October 16. Three changes deserve a look before you upgrade: a new setting picks the model behind client-side auto mode, subagent defaults moved outside the terminal, and hooks now run at most 12 at a time per event.

This brief picks up after the 2.1.299-2.1.300 notes. It is not the full list.

This site is independent and not affiliated with Anthropic. We read the changelog and the docs. We did not install or run these versions, so nothing below is a hands-on result.

A model setting for the auto mode classifier

In auto mode, Anthropic's permission modes page says "a second model, the classifier, reviews actions instead of you". The 2.1.302 entry adds a way to choose that model:

Added CLAUDE_CODE_AUTO_MODE_MODEL to choose the model the client-side auto mode classifier uses; it does not change the model for server-side auto mode

The entry names no default and no list of accepted values. Treat the setting as untested until you have tried it on a throwaway project. Our guide on stopping auto mode from acting without confirmation covers how to leave auto mode altogether.

2.1.301 also adds telemetry for people who export it:

Added classifier_verdict and permission_mode to OpenTelemetry tool_decision events, so auto mode classifier decisions can be told apart from permission rules

Subagent defaults change outside the terminal

The docs define a fork this way: "A fork is a subagent that inherits the entire conversation so far instead of starting fresh." The 2.1.302 entry turns forks on in more places:

Changed fork subagents (subagent_type: "fork") to be on by default in claude -p, the Agent SDK, the desktop app and VS Code, as they already are in the terminal; CLAUDE_CODE_FORK_SUBAGENT=0 turns them off, and claude mcp serve never offers them

A second entry changes where subagents run:

Changed the desktop app and VS Code to always run the main conversation's subagents in the background, as the terminal does, rather than let Claude choose; a subagent's own Agent calls in claude -p, the Agent SDK, the desktop app and VS Code now run in the foreground by default

Our reading: if you script claude -p or build on the Agent SDK, a task that used to run in one context may now spawn forks. Check your scripts after upgrading. For the basics, see our subagents guide.

Hooks wait their turn

The 2.1.302 entry on hooks is one line:

Changed prompt and agent hooks so that at most 12 of one event run at a time; the rest wait their turn

The entry says "one event", not all hooks combined. If you run many prompt or agent hooks on the same event, expect the extra ones to queue. Command hooks are not named in the entry. Our guide on making hooks fail closed covers the blocking side.

The Claude moves worth stealing, by email

Workflows tested on real jobs, sent only when they are worth it. Free.

Subscribe

Idle compaction gets a switch in /config

The 2.1.299 release added idle compaction. In 2.1.302 it becomes easier to find:

Added an Idle compaction row to /config for turning idle compaction off or back on; the notices shown after an idle compaction now say how to turn it off

You no longer need to edit settings.json by hand. The earlier brief explains what idle compaction does and why you might turn it off.

Unattended sessions and third-party providers

Two entries matter if you run Claude Code without watching it. From 2.1.301:

Added CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS so unattended retry mode (CLAUDE_CODE_RETRY_WATCHDOG) ends a request at a usage limit that resets too far away instead of waiting

And another entry in this pair of releases:

Changed claude -p and Agent SDK sessions with no permission mode to start in default mode on third-party providers, with telemetry off or on a fresh install; --permission-mode auto keeps auto

The first caps how long a stalled run waits. For limits on a normal plan, see our rate limits guide. The second is a safer starting mode, but a script that relied on auto will now need the flag.

Smaller changes worth a glance

• Claude Design: /design-sync and /design-login now show a notice that standalone Claude Design (claude.ai/design) closes by December 14, 2026.
• Mods: 2.1.301 adds rows to a mod's Input, and changes when a mod hook's .catch handler runs. Our Claude Mods guide covers enabling them.
• VS Code: 2.1.301 adds LaTeX math rendering in Claude's responses and a "Get more usage" link in the usage-limit banner for Pro and Max plans.
• Footer hint: the ← for agents hint hides 10 seconds after a session opens.
• Cloud sessions: a routine's page now shows a banner saying why its last run did not start.

Upgrade readout

If you	Do this
Use auto mode and want a different classifier	Try CLAUDE_CODE_AUTO_MODE_MODEL on a test project first
Script claude -p or the Agent SDK	Re-run your scripts: forks are on by default, CLAUDE_CODE_FORK_SUBAGENT=0 turns them off
Run many prompt or agent hooks	Expect more than 12 of one event to queue
Leave long terminal sessions idle	Check the new Idle compaction row in /config
Run unattended retries	Look at CLAUDE_CODE_RETRY_WATCHDOG_MAX_USAGE_LIMIT_WAIT_MS
Anyone else	Routine upgrade

What this brief does not cover

The changelog lists more entries under both versions than we summarize here, including several for Claude Tag admin settings. We did not read the docs for every setting named above, and we ran none of them. Anything before 2.1.301 is outside this note.

What is CLAUDE_CODE_AUTO_MODE_MODEL in Claude Code?

An environment variable added in 2.1.302 that chooses the model the client-side auto mode classifier uses. The changelog says it does not change the model for server-side auto mode. It names no default and no accepted values.

Are fork subagents on by default in Claude Code now?

Since 2.1.302, yes in claude -p, the Agent SDK, the desktop app and VS Code, as the changelog says they already were in the terminal. Set CLAUDE_CODE_FORK_SUBAGENT=0 to turn them off. claude mcp serve never offers them.

How many hooks can Claude Code run at once?

For prompt and agent hooks, at most 12 of one event run at a time as of 2.1.302. The rest wait their turn. The changelog entry does not mention command hooks, so we cannot say whether the same cap applies to them.

Sources (checked 2026-10-17)

• Claude Code changelog: entries for 2.1.301 (October 15, 2026) and 2.1.302 (October 16, 2026).
• Claude Code subagents docs: definition of a fork.
• Choose a permission mode: the auto mode classifier.