Claude Code 2.1.304 to 2.1.306: resume warnings, wider auto mode setup

Update Claude Code to 2.1.306. On a Pro or Max plan you now get a warning before resuming an inactive conversation would use a noticeable share of your 5-hour usage limit, and /auto-mode-setup now runs on Amazon Bedrock, Google Vertex AI, Microsoft Foundry, the Claude apps gateway and telemetry-off sessions, per the CHANGELOG. If you use a cloud provider or gateway, don't stop at 2.1.304: the CHANGELOG says 2.1.305 fixed a 2.1.304 regression that could make every request fail when the endpoint or gateway rejects tool search.

What changed#

Three releases reached npm in two days: 2.1.304 at 3:47 PM ET Monday, 2.1.305 at 3:29 PM ET Tuesday and 2.1.306 at 10:28 PM ET Tuesday. When we checked npm at 2:45 PM ET Wednesday, latest and next pointed to 2.1.306 and stable to 2.1.304.

	Before	Now
Resuming an idle conversation (Pro, Max)	No warning	A warning with two choices: resume anyway, or start fresh and have Claude look up the old conversation (2.1.304)
/auto-mode-setup	Not in sessions with feature-flag fetching off	Available on Bedrock, Vertex, Foundry, the gateway and telemetry-off sessions (2.1.304)
/usage warnings	Plain warnings	Each warning is a one-line fix; press f to apply (2.1.305)
Exiting claude --worktree with a detached or switched HEAD	Could delete the worktree and branch, which could lose commits	Fixed (2.1.304)

2.1.304 also lists the Code Review levels (Quick, Standard, Deep) that we covered on Oct 16, and adds /claude-api managed-agents-secure, a read-only check of a Managed Agents agent against Anthropic's security guide. 2.1.306 fixes an API error showing up as "prompt is too long" in sessions with many MCP tools.

Why it matters#

Resuming has a hidden cost. The caching docs say that on resume "Claude Code sends the whole conversation again," and after a long gap "the next request recomputes the full input." The warning now tells you before that happens. Auto mode setup matters on cloud providers because auto mode stops itself when blocked: per the permission modes page, after 3 blocks in a row or 20 in total, "auto mode pauses and Claude Code resumes prompting."

How to use it#

1. Run claude update (or npm i -g @anthropic-ai/claude-code@2.1.306), then claude --version.
2. When the warning appears on an old thread, start a new conversation and use this prompt (ours, untested):

Look up my previous conversation in this project about <topic>. Summarize where we left off: decisions made, files changed, and open TODOs. Don't run or edit anything yet.

3. If auto mode keeps stopping to ask, run /auto-mode-setup. Per the auto mode docs, it reads your project's CLAUDE.md, README.md, config files and git remotes, plus hosts and command names from recent sessions, "never your messages." Then run claude auto-mode config.

Good to know#

• The stable channel has the 2.1.304 bug. 2.1.305 fixed "every request failing on Vertex, Bedrock and other providers when the endpoint or a gateway rejects tool search," calling it "a 2.1.304 regression." As of 2:45 PM ET Wednesday, npm's stable tag still points to 2.1.304. Our read: if you follow stable on a provider or gateway, install 2.1.305 or later by version number.
• It is all or nothing. The docs say: "You accept or discard it as a whole, so edit ~/.claude/settings.json afterwards to adjust single entries."
• The trigger is not defined. The CHANGELOG says only "a noticeable share" and gives no number.

What to watch: when stable moves past 2.1.304, and whether Anthropic publishes the threshold behind the resume warning.