Claude Code 2.1.303-2.1.304: auto mode setup, review levels

Claude Code 2.1.304 is dated October 19, 2026 in Anthropic's changelog. Version 2.1.303, dated October 18, lists only "Bug fixes and reliability improvements". Three changes in 2.1.304 deserve a look before you upgrade: a guided /auto-mode-setup, review levels for Code Review, and a warning before you resume an old conversation on Pro or Max.

This brief picks up after the 2.1.301-2.1.302 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.

Auto mode gets a guided setup

Anthropic's permission modes page says that in auto mode "a second model, the classifier, reviews actions instead of you". The 2.1.304 entry adds a command that teaches that classifier about your project:

Added /auto-mode-setup: a guided setup that reads your project and recent sessions to teach auto mode about your environment, now also available on Bedrock, Vertex, Foundry, Claude apps gateway and telemetry-disabled sessions

The entry does not say what the setup writes or where it stores the result. The permission modes page we read does not mention the command either. Try it on a throwaway project first. To leave auto mode instead, see our guide on stopping auto mode from acting without confirmation.

A warning before you resume an idle conversation

If you are on Pro or Max, the changelog adds a check before you pick an old session back up:

Added a warning on Pro and Max plans when resuming an inactive conversation would use a noticeable share of your 5-hour usage limit: resume anyway, or start a new conversation and Claude can look up the previous one if you refer to it

"Noticeable share" has no number attached, so you cannot predict when the warning appears. The practical takeaway is the second option: a fresh conversation that refers to the earlier one. Our rate limits guide covers what else eats into the window.

Code Review gets three levels

Two entries tagged Code Review land in 2.1.304:

Added review levels to Code Review: Owners choose Quick, Standard (the default), or Deep for the organization, and can give individual repositories their own level

Added an option in Code Review admin settings to review every repository in a GitHub organization, including ones added later, without selecting them one by one

Anthropic's Code Review docs describe Quick as "a lighter pass. The fastest level, at the lowest cost." The same page lists Deep as "the most thorough pass. Takes the longest and costs the most."

Code Review is for Team and Enterprise plans, so this matters to admins more than to solo users. Our Code Review launch brief covers what the service does.

The Claude moves worth stealing, by email

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

Subscribe

A security comparison for Managed Agents

One new slash command points at Anthropic's published security advice:

Added /claude-api managed-agents-secure, which compares a Managed Agents agent's settings with Anthropic's published security guide, without changing anything

The guide it refers to is the Managed Agents secure deployment page, which "lists the settings that limit what an agent on Claude Managed Agents can reach and do, most important first." We did not run the command, so we cannot say how detailed its output is. Our secure deployment brief walks through the page itself.

Mods get more control

Several 2.1.304 entries extend what a mod can do. Two stand out for anyone writing or managing them:

Added decider: "person" to a mod's tool.check answer: only the permission dialog or a host's callback can approve the ask, and it is refused with the mod's reason where nobody can be asked

Changed mods' settings.read hooks to receive each settings layer and rewrite the merged settings, so a managed mod can make an organization's setting a ceiling, including the cleanupPeriodDays the retention sweep uses

The first lets a mod insist that a human approves a call. The second gives a managed mod a way to cap a setting for a whole organization. Both are written for mod authors. If you only want to turn mods on, start with our Claude Mods guide.

Compaction and skills

Compaction changes in two small ways. The compaction summary in the transcript view (Ctrl+O) now lists the names of tools called earlier. After a compaction, Claude is sent the skill list again:

Changed compaction to send Claude the skill list again afterwards; /reload-skills now sends only new or changed skills

Skill hooks also get a ceiling: "the skills of one source (a repo, your own folder, or a plugin) can register at most 1,024 hooks and 16,384 matcher characters per session". You are unlikely to hit it with a handful of skills.

Smaller changes worth a glance

• Tool search: now on by default behind a custom ANTHROPIC_BASE_URL, and a 400 or 422 from the gateway is retried without it.
• Artifacts: the /artifacts delete confirmation shows numbered choices. 1 or Enter keeps the artifact, 2 deletes it.
• Fast mode: the "(research preview)" label is gone from the /fast dialog title on Opus 5.5.
• VS Code: background sub-agents show as one status line under their Agent card, and the full transcript stays in /tasks.
• Plugins: claude plugin validate no longer prints notes predicting where a mod's hooks would run.

Upgrade readout

If you	Do this
Use auto mode in a real project	Try /auto-mode-setup on a test project first
Resume old sessions on Pro or Max	Read the new warning, and consider a fresh conversation
Administer Code Review	Pick a level per repository: Quick, Standard or Deep
Run Managed Agents	Try /claude-api managed-agents-secure, it changes nothing
Write or manage mods	Read the decider and settings.read entries
Anyone else	Routine upgrade

What this brief does not cover

The changelog says "12 improvements" and "43 bug fixes" sit under expandable sections for 2.1.304, and we did not read them. We did not run any command named above. Anything before 2.1.303 is outside this note, and 2.1.303 itself carries a single line.

What is /auto-mode-setup in Claude Code?

A command added in 2.1.304. The changelog calls it "a guided setup that reads your project and recent sessions to teach auto mode about your environment". The entry does not say what it stores, so test it on a throwaway project first.

What are the Code Review levels in Claude Code?

Quick, Standard and Deep. Standard is the default. Per the changelog, Owners choose a level for the organization and can give individual repositories their own. Anthropic's docs describe Quick as the fastest and cheapest, and Deep as the most thorough and costliest.

Why does Claude Code warn before resuming a conversation?

Since 2.1.304, Pro and Max plans see a warning when resuming an inactive conversation would use a noticeable share of the 5-hour usage limit. You can resume anyway, or start a new conversation and refer to the previous one.

Sources (checked 2026-10-20)

• Claude Code changelog: entries for 2.1.303 (October 18, 2026) and 2.1.304 (October 19, 2026).
• Choose a permission mode: the auto mode classifier.
• Code Review: review levels.
• Secure your Managed Agents deployment: the guide the new command compares against.