What happened
Anthropic published Claude Code v2.1.318 on GitHub on November 4, 2026. The release adds an effort option to a mod’s $.model.fork. A side job can therefore run at a lower effort level than the surrounding session; the result reports the effort sent. This is a scoped control for delegated work rather than a change to the model or to the overall session’s effort setting.
The Claude apps gateway now supports assume_role on Mantle upstreams. The release says this allows a role in another AWS account and per-developer AWS cost attribution. It does not describe additional configuration steps or quantify savings, but the capability matters to teams whose gateway traffic needs to be associated with distinct AWS identities or accounts.
The release also exposes the expiry date of a stored claude.ai sign-in. claude auth status reports it as loginExpiresAt, while JSON output from claude -p uses credential_expires_at. In text output, claude -p writes a one-line warning to stderr beginning 10 days before the stored sign-in expires. This gives scripts and operators a way to inspect or notice impending expiry before it interrupts work; the notes do not say that the tool renews credentials automatically.
The remaining changes are largely reliability fixes. The release fixes a regression in 2.1.314 where --continue could lose earlier conversation after retries for a malformed tool call ran out, and corrects idle compaction that could replace a conversation with a reply that was not a summary. For MCP, claude mcp add and remove now wait for the config lock or exit with code 1, although another job’s save can still replace a saved server. Stdio MCP servers that exit on the version probe are no longer re-probed on every start; a server that dies in under three seconds for its own reason is also pinned to the legacy handshake for seven days.
Other fixes touch cloud sessions, remote-control hosts, self-hosted runners, plugins, permission prompts, login state, and non-interactive operation. The notes include issues such as jobs reporting “Not logged in” after another terminal or host signs in to the same account and organization, and cloud sessions stopping mid-turn when uploads are refused for several minutes. No price change, new general availability, or access restriction is announced.
What the archive adds
This is a new release in an ongoing Claude Code update thread, not a restatement of the prior changes. Claude Code v2.1.310 covered Monitor parity across Bedrock, Vertex and Foundry, plus gateway retry and Enterprise role controls; Claude Code 2.1.314 covered idle compaction for long conversations in VS Code and Claude Desktop. This release adds different controls and fixes.
Trend signal
The release weakly supports agent economics moves to completed work: per-fork effort and per-developer AWS cost attribution are controls relevant to managing the resource use of agent work. It does not provide cost-per-task measurements, usage analytics, or evidence that completed-task economics are replacing token or call metrics, so the connection is suggestive rather than conclusive.
Why now
The release specifically adds visibility into stored claude.ai sign-in expiry and a warning beginning 10 days before it. That timing makes the change operationally useful for teams that might otherwise discover expiry only when a session or automated job can no longer authenticate.
Connections to other events
An earlier Claude Code release extended idle compaction to VS Code and Claude Desktop, while v2.1.318 fixes a case where idle compaction could produce a non-summary. Together, these changes show the feature’s practical reliability is still being refined. An earlier Windows utility, mcp-restart, addressed changing MCP servers within a running session; v2.1.318 instead focuses on config locking and server startup behavior.
Why it matters
The most immediately useful addition for operators is credential-expiry visibility. A warning with a ten-day lead gives teams time to update sign-in procedures before interactive users or scheduled claude -p jobs encounter an expired credential. JSON fields also make expiry easier to inspect programmatically, though the release does not promise automatic renewal.
For teams using mods, lowering effort for a side job can help avoid spending the session’s full effort level on a smaller task. The gateway’s cross-account role support and per-developer attribution can make AWS usage easier to associate with owners. Neither feature comes with measured savings, so teams should treat them as controls and observability improvements, not proof of lower costs.
The MCP locking fix is relevant where multiple processes modify the same configuration: waiting for a lock reduces the chance of conflicting writes. The release explicitly retains a limitation—another job’s save can still replace a saved server—so concurrent configuration is not made fully transactional. The broader list of fixes is useful to users affected by those specific failures, but it does not establish a general performance or reliability gain for every Claude Code workflow.
What happens next
- Teams using headless Claude Code will check credential expiry in automation or operational runbooks. (3 months, 65%). Watch for: scripts or documentation adopting
credential_expires_ator the stderr warning.
Limits and uncertainty
The supplied release notes do not give benchmarks, adoption data, or examples quantifying the cost effect of lower fork effort or per-developer AWS attribution. They do not explain the full setup for assume_role, and they do not establish that credentials can be refreshed automatically. The release lists many fixes but does not quantify how often the underlying failures occurred or their impact. The MCP config change adds locking, yet the notes explicitly say another job’s save can still replace a saved server. These claims should be read as release-note descriptions of changed behavior, not independent verification of outcomes across deployments.
Skill to build now
If you run Claude Code headlessly, add a check for credential_expires_at from claude -p JSON output to a scheduled health check or runbook, and alert before the stored sign-in expires. For mod-based workflows, try setting a lower effort on a genuinely secondary fork and compare whether the result is sufficient for that job. The release provides the interfaces, but not a prescribed threshold or a measured cost-saving figure.
Reader takeaway
Claude Code v2.1.318 is a targeted operational release: it adds effort control for mod forks, AWS role and cost attribution support, and advance visibility into sign-in expiry, while fixing MCP and session edge cases. Teams should use the new expiry field and warning where authentication failures would interrupt work, and treat the other controls as workflow-specific rather than broad performance promises.