Claude Code Digest — 2026-03-23 20:16:35
What the docs reveal
Anthropic is pushing Claude Code beyond interactive chat. The latest documentation changes reveal a definitive shift toward autonomous background execution. Developers can now schedule tasks to run on Anthropic's managed cloud, local desktops, or CI/CD pipelines. This transforms Claude from an on-demand coding assistant into a persistent daemon.
The new /schedule command handles cloud task creation conversationally. You tell Claude what to run and when, and it configures the infrastructure. Anthropic clearly sees managed execution as a strategic growth area. Hosting the agentic loop keeps your tasks tethered to their managed cloud, abstracting away the boilerplate of cron jobs and state management. For developers, this allows frictionless deployment of continuous tasks like repository monitoring, dependency updating, and automated refactoring.
Runaway agents cost money. To mitigate risk, Anthropic added a strict three-day expiration limit for all recurring tasks created in Claude Code. Without a human to hit stop, a looped prompt could easily burn through thousands of tokens or spam downstream APIs. This three-day circuit breaker forces developers to re-authorize long-running jobs. It protects Anthropic's infrastructure from zombie processes and saves users from catastrophic billing surprises.
The documentation also clarifies task durability boundaries. The docs now explicitly warn that in-process /loop commands die immediately when the terminal session ends. Users likely flooded Anthropic with complaints about unexpected task terminations after closing their laptops. To solve this, a new workflow automation table guides developers toward the right tool. It pushes users toward managed cloud or GitHub Actions for anything requiring persistence. If you need a task to survive a reboot, you must deploy it outside your active terminal.
Running prompts while you sleep dictates a new way to write them. You cannot answer clarifying questions when you are offline. Consequently, Anthropic added stringent best practices for autonomous prompting. Your prompts must now resemble strict function contracts rather than informal dialogue. Developers must define explicit success criteria and mandate exact result handling methods, such as writing to a log file or opening a pull request. An ambiguous prompt in a scheduled task will stall or hallucinate.