This week in Claude Code, Codex and Gemini CLI (week of November 8, 2026)

The change with the widest reach this week is that auto mode no longer falls back to the built-in classifier. Where telemetry or feature flags are off, including behind an LLM gateway, calls the server cannot check are asked about or denied.

For unattended runs, permissions.denyInsteadOfPrompting is new: instead of stalling on a prompt, the call is denied and the turn continues.

Claude Code

Three releases: 2.1.317, 2.1.318 and 2.1.319.

Auto mode lost its fallback (2.1.319)

The CHANGELOG reads: "Changed auto mode with telemetry or feature flags off, including behind LLM gateways: calls the server can't check are asked about or denied, with no fallback to the built-in classifier; set CLAUDE_CODE_AUTO_MODE_SERVER=0 to keep using it (Bedrock, Vertex and Foundry are unchanged)."

Environment	Effect
Telemetry or feature flags off	Calls the server can't check are asked about or denied
Behind an LLM gateway	The same
Bedrock / Vertex / Foundry	Unchanged

CLAUDE_CODE_AUTO_MODE_SERVER=0 restores the previous behaviour of using the built-in classifier. In the other direction, CLAUDE_CODE_AUTO_MODE_SERVER=1 turns the server-side check on behind a gateway even with CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS set — and where managed settings set the latter, they must set the 1 too.

Relatedly, /status no longer reports "Auto mode server: Enabled" when the server-side review is not actually reaching the session. For how the modes differ, see Claude Code starts in auto mode by default.

New settings and variables (2.1.319)

Setting	What it does
permissions.denyInsteadOfPrompting	Denies a call that would wait on a permission prompt and continues the turn, in terminal sessions, subagents and teammates
CLAUDE_CODE_DISABLE_FEATURE_FLAGS	Stops feature-flag fetching without turning telemetry off; every flag uses its built-in default
CLAUDE_CODE_DISABLE_MCP_TASK_BACKGROUND	Waits for an MCP task inside its tool call instead of backgrounding it immediately

MCP tasks are now supported in the terminal: when a server runs a tool call as a task, Claude Code moves it to the background right away and reports the result when it finishes.

A check before your email address goes out (2.1.317)

New in 2.1.317: "Added a check that asks before a web or shell tool call sends your account email address to the network."

Three security fixes also landed:

• A sed command could skip the permission check in accept-edits and auto mode (2.1.319)
• Bash deny and ask rules missed commands written with quoted words (2.1.319)
• /rewind could restore or delete a file through a directory symlink (2.1.319)

Sign-in expiry is now visible (2.1.318)

This one matters for unattended runs. The expiry date of a stored claude.ai sign-in is now available as loginExpiresAt in claude auth status and credential_expires_at in claude -p JSON results, with a one-line stderr warning starting 10 days before.

The release itself is mostly fixes for configuration files and stored logins being corrupted on network home directories, so it is worth taking if you run on a shared home. Claude Code also no longer installs or updates the IDE extension when started from an IDE terminal; use /ide from an outside terminal instead.

Codex

v0.167.0 ships six features.

Addition	What it does
codex agents --no-daemon	Opens the agent command center; it and the session picker now mark unread results
Backend switching	Switch between connected backends from the terminal session picker, keeping drafts
Image preview	Preview local image attachments while composing, on supported terminals
gcp_setup_wizard	Sets up Google Cloud Vertex AI with Application Default Credentials or a credential file
System proxy	Opt into macOS or Windows system proxy settings from /experimental
Patch details	Expand patch details in the transcript while an approval prompt is open

gcp_setup_wizard continues from v0.165.0, which made Vertex AI a built-in provider via model_provider = "gcp" — see Codex CLI v0.165 makes Vertex AI a built-in provider. The API key and cloud provider options moved under the "Sign in another way" menu.

Three fixes stand out. Saved chats stay visible after switching model providers, and a chat that cannot resume can still be read and turned into an editable prompt for a new chat. Network approval prompts close when their requests end, and concurrent requests to the same endpoint prompt one at a time. Automatic context compaction now accounts for retained output from failed responses.

Gemini CLI

v0.67.0 (November 4) has two features among 27 entries: the models command, and the fix for a Claude hook timeout that migrated 1000x too small. See Gemini CLI v0.67 adds a models command and fixes the unit in Claude hook migration.

What to check now

1. If you run auto mode behind an LLM gateway, note the 2.1.319 change and set CLAUDE_CODE_AUTO_MODE_SERVER if needed
2. For unattended runs that must not stall, consider permissions.denyInsteadOfPrompting
3. Check your sign-in expiry with claude auth status
4. Move to 2.1.318 if you run on a network home directory

Summary

• Auto mode asks about or denies calls the server can't check, with no classifier fallback
• CLAUDE_CODE_AUTO_MODE_SERVER=0 restores the old behaviour; Bedrock, Vertex and Foundry are unchanged
• permissions.denyInsteadOfPrompting turns a stalled prompt into a denial
• Web and shell tool calls ask before sending your account email address
• Codex 0.167.0 adds gcp_setup_wizard and opt-in system proxy settings