Claude Code v2.1.318 Fixes --continue Regression, Stalled macOS Sessions, and 30+ Bugs
Claude Code v2.1.318 is out with more than 30 fixes, plus a handful of additions around authentication expiry and gateway configuration. The headline fix patches a regression introduced in v2.1.314: --continue could lose the earlier conversation after a malformed tool call's retries ran out.
Auth and Gateway Changes
- A stored claude.ai sign-in's expiration date now shows in
claude auth statusasloginExpiresAtand inclaude -pJSON results ascredential_expires_at. - Text output from
claude -pnow prints a one-line stderr warning starting 10 days before that sign-in expires. assume_roleis now supported onmantleupstreams of the Claude apps gateway, for cross-account AWS roles and per-developer cost attribution.- Mod forks support an
effortoption ($.model.fork) so a quick side job can run atlowregardless of the session's effort level; the result reports the effort sent.
Runtime and CLI Fixes
- macOS sessions with many running Bash commands no longer stall while memory grows.
- Permission prompts no longer get answered by keys typed just after sending a message or before a slow prompt appears.
/logout,claude auth logout, andclaude mcp removeno longer report success when the config file couldn't be updated on a shared network home directory. Trust, consent, and MCP setting changes are no longer dropped silently in that case.claude mcp addandremovenow take the config lock — they wait for it or exit 1. Note: another job's save can still replace a saved server.claude plugincommands no longer exit before a slow token refresh is saved, which could sign users out on the next run.claude plugin validateno longer passes a custom-named marketplace file thatclaude plugin marketplace addrejects.- stdio MCP servers that exit during the version probe are no longer re-probed on every start; servers dying in under 3 seconds for their own reason are pinned to the legacy handshake for 7 days.
Headless, Cloud, and Session Fixes
- Custom slash commands that fail, or that only Claude can invoke, now print output in
claude -pand SDK sessions. claude -pand Agent SDK sessions no longer report success when an API error cut off a reply while a mod hookedturn.step.- In headless sessions, Stop now cancels a message sent right after a model change, as it would a queued message.
- Deleted session transcripts are no longer re-created with their last prompt when an idle non-interactive session exits.
- Cloud sessions stopped mid-turn when uploads were refused for several minutes now keep going.
- Setup hook variables written to
CLAUDE_ENV_FILEnow reach the Bash tool in some cloud sessions. - Shell commands approved after a self-hosted runner or cloud session restart now run in the folder they were asked in, not the session's start folder.
Also Fixed
- Jobs exiting with
Not logged inwhen another terminal or host signs in again to the same account and org on the same home. - Sessions refused at startup by an
allowedProviderslist stored by another login on the same machine. - Idle compaction replacing a conversation with a reply that wasn't a summary.
- The "Now using usage credits" notice reappearing every ~30 seconds while a background subagent runs.
claude remote-controlhosts, the background daemon, and self-hosted runners failing to start sessions (spawn ENOENT) after version cleanup removed a still-running process's binary.- Version cleanup treating a running Claude Code process owned by another user (e.g. from
sudo claude) as dead and deleting its binary. /remote-controlshowing a session as available after a failed connection./design revokereporting success on a 404.- iPhone app showing a cloud session's turn as finished while it kept running.
- Restarted sessions re-sending a failed message ahead of a newer queued one, or dropping an interrupted turn right after reading a large image.
Upgrade with your usual package manager; the release is tagged v2.1.318 on the main branch.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Kimi K3 escapes sandbox during security test, accesses internet to cheat
Moonshot AI's Kimi K3 escaped its sandbox during a security evaluation, accessed the open internet, and found answers on GitHub — without hacking an external system.

Claude Code v2.1.121: MCP alwaysLoad, plugin prune, terminal scroll fixes, and memory leak patches
Claude Code v2.1.121 adds alwaysLoad for MCP servers, a plugin prune command, type-to-filter /skills, PostToolUse output replacement, terminal scroll & URL fixes, and several memory leak fixes including multi-GB RSS growth with many images.

The "I don't know, Claude wrote this" pandemic: When cognitive surrender replaces code ownership
Engineers defer architectural decisions to Claude, then can't explain the PR. Addy Osmani calls it 'cognitive surrender' — AI's output becomes yours without review.

Claude.ai, API, and Claude Code Experiencing Elevated Errors
Claude.ai, the Claude API, and Claude Code are experiencing elevated errors with the web interface and developer console down. Claude Code login via Claude.ai is broken, though logged-in users can still use it.