Claude Code adds scheduled cloud tasks for PR reviews and `/schedule` runs

• Anthropic's scheduled tasks docs add recurring cloud-run prompts to Claude Code, so repo work like daily PR review and overnight CI analysis can keep running when your machine is off.
• Rohan Paul's PR auto-fix post shows the adjacent workflow change: Claude Code can stay attached to a GitHub pull request, react to failed checks or review comments, and push "obvious fixes" back to the branch from Anthropic-managed infrastructure.
• The new scheduler is exposed in the web app, desktop app, and the /schedule CLI flow, and Anthropic's docs say task setup includes a named prompt plus an environment selection that scopes repo and connector access.
• Early Hacker News discussion around the launch thread points to practical constraints, including plan-level caps, no screenshots, and restricted outbound network access to only some domains.

Anthropic has moved a slice of Claude Code from "agent on your laptop" to agent jobs running on its own cloud. In the product docs, scheduled tasks are described as recurring prompts that run on Anthropic infrastructure, with examples including "daily PR reviews" and "overnight CI analysis." The same docs say the feature is available on Pro, Max, Team, and Enterprise, and can be created from the web UI, desktop app, or with /schedule in the CLI via the docs.

The practical effect is visible in Rohan Paul's demo thread, which describes Claude Code as a background PR mechanic that can "watch a GitHub pull request," respond to CI failures and review comments, clone the repo, run tests, edit code, and push fixes after you close the app. That goes beyond reminder-style scheduling: the cloud worker can keep state around a repo task and decide whether a reviewer note is "clear enough to fix on its own" or should be handed back to the developer.

The launch is useful, but the first wave of practitioner feedback is about boundaries. The Hacker News thread highlights a plan cap of "3 daily cloud scheduled sessions" for at least some users, plus open questions about whether cloud environment runtime itself is billed separately from tokens. Those details matter because the feature is positioned for recurring CI and review loops, where frequency and unattended runtime determine whether it fits real team workflows.

Users in the same discussion summary also call the environment "a bit restrictive": one commenter says it "doesn't take screenshots," while another says it "doesnt allow egress curl, apart from few hardcoded domains." That makes the first release more suitable for repo-bound automation and approved integrations than for arbitrary browser or network-heavy jobs, even though Anthropic's docs say environments can connect to repos and MCP tools like Slack and Linear.