Claude Code cloud sessions launch, credit expires October 7

Anthropic made Claude Code cloud sessions generally available on September 23, per AlphaSignal. That matters for anyone near Pro or Max usage limits, since background jobs no longer compete with interactive work for the same quota.

Sessions run on Anthropic-hosted VMs, so a task keeps going after you close the laptop. Each one works on its own git branch, started from claude.ai/code, the mobile Code tab, the desktop app, or claude --cloud.

Existing subscribers get a one-time credit, $100 on Pro and $250 on Max, claimed with /claim-credit or at claude.ai/code/claim-credit by October 7. It requires a linked GitHub account, spends before your normal plan usage, and only applies to cloud sessions. Two weeks is tight for a deadline tied to setup work, so claim it now and pick a task later.

The best fit is expensive, long-running work, dependency migrations, test suite repairs after a framework bump, the six-hour refactor you've put off for four months. Those are per-branch tasks you can check afterward, the shape cloud sessions are built for.

The limit to design around is that each session runs on a VM you don't watch. A local credential, a network-only service, or a judgment call mid-run all make a session a poor fit. A test suite that verifies the result on its own makes it a good one.

A September 23 writeup on the bug found that Claude Code skips AGENTS.md entirely when telemetry is disabled, confirmed with a canary-word test. The loader added in version 2.1.277 sits behind a remote flag that defaults to false, and setting DISABLE_TELEMETRY=1 drops your AGENTS.md through at least 2.1.280. GitHub issue #95690 is still open, and the workaround is a one-line CLAUDE.md containing @AGENTS.md. The result is a silently degraded agent for anyone who disables telemetry, with no warning that instructions were skipped.