Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Coding tasks

Claude Code 2.1.318: sign-in expiry and fork effort

Written November 4, 2026

Sign-in expiry is now surfaced before it breaks a session, and forked subagents take an effort setting. What to check after upgrading.

What changed

Claude Code 2.1.318 fixes the quietest way a session dies: a stored claude.ai sign-in reaching its time limit. The changelog adds the date a stored sign-in ends to claude auth status as loginExpiresAt, and to claude -p JSON results as credential_expires_at. Headless runs get a one-line warning on stderr, starting 10 days before the sign-in ends. And when a sign-in does reach its limit, the failure message now says so; in claude -p it names claude auth login and claude setup-token as the way back in. Previously, a session could simply stop working with no warning that expiry was the cause.

The release also adds an effort option to a mod's $.model.fork, so a quick side job forked from a session can run at low effort whatever effort the session itself uses. The result reports the effort that was sent, so a fork's cost level is visible rather than assumed. For the difference between forked subagents and parallel subagents, see Run parallel subagents in Claude Code.

Three further changes affect shared and cloud setups. Cloud sessions gain a No repository option: a session on an Anthropic-hosted environment can start without picking a repository. Saves of settings files and the stored login on shared network home directories are more reliable — less often deleted, briefly emptied, or replaced by a near-empty or older copy — with matching fixes for cases where a failed config write was reported as success. And the IDE extension is no longer installed or updated when Claude Code starts inside an IDE terminal; install or update it from /ide in an outside terminal or from the editor's extensions view.

Two changes close gaps that could silently widen access. Entries in deniedMcpServers given as URLs now also block Anthropic-provided servers whose upstream URL matches, and connectors in cloud sessions and on self-hosted runners where the session is given the connector's own URL; in other cases, deny a connector by name. And the Claude apps gateway gains assume_role on mantle upstreams, for assuming a role in another AWS account and for per-developer AWS cost attribution. The release also records the answer when Claude Code asks whether to turn on the mods it writes in a session. A regression from 2.1.314 is fixed: --continue sometimes lost the earlier conversation after a malformed tool call's retries ran out. The release carries 47 bug fixes in total; the rest are not listed here.

After upgrading, check

  1. Look at your sign-in expiry date. Run claude auth status and find loginExpiresAt. If the date is within days, sign in again before a long unattended run depends on it. If you parse claude -p JSON output in scripts, look for credential_expires_at under the same meaning, and watch stderr for the 10-day warning in text output.
  2. Set fork effort where a side job should stay cheap. If a mod forks a quick side job from a high-effort session, set the fork's effort to low in its $.model.fork setting, and confirm the reported effort in the result. A fork without an explicit setting inherits the session's effort level.
  3. Check denylist entries given as URLs again. If you blocked a server by URL and expected Anthropic-provided servers with the same upstream URL to be blocked too, they now are. If you relied on the old behaviour — URL entries blocking only the named server — tighten or re-test the entry, because the block is now wider.
  4. Fix IDE extension installs that ran from an IDE terminal. If your setup installed or updated the extension by starting Claude Code inside an IDE terminal, that path no longer runs. Use /ide from an outside terminal or the editor's extensions view instead.
  5. On a shared network home directory, watch one save cycle. The reliability fixes target saves being deleted, emptied, or replaced by older copies. After upgrading, make one settings change and one sign-in-dependent action, then confirm the files on disk kept the change. A single clean cycle does not prove the old failure is gone for every filesystem, but a lost change on the first cycle says the problem is still present on yours.
  6. Cloud-session users: try starting without a repository. If a cloud session exists only to run a task that needs no repo, use the No repository option instead of picking a placeholder repository. Sessions that need a repo are unchanged.

Limits

Related playbooks

Sources

Back to: Coding playbooks hub · Also: Claude Code 2.1.317: schedule warnings and cost reports · Run parallel subagents in Claude Code · Baseline usage before a shared allowance