Automations

Workflows that run without you

A guide to the Claude workflows that run without you in the loop — hooks that fire when events happen, tasks that run on a schedule, and pipelines that carry Claude’s output into the next step without anyone driving. Drawn from production systems that have been running unattended for months.

Five areas, one automated system

• hooks — Event-triggered Claude tasks using the four hook types: PreToolUse, PostToolUse, Stop, and SubagentResult. Each hook is a shell command or Python script; non-zero exit from PreToolUse blocks the tool call. Learn the failure modes that only surface at 2 a.m.
• scheduling — Running Claude jobs on a clock: cron syntax (0 6 * * 1 for Monday 6 a.m.), Windows Task Scheduler, systemd timers on Linux. Non-interactive mode via claude --no-interactive --print, stdout captured to a date-stamped log file, exit code checked by the scheduler.
• pipelines — Chaining Claude calls so each step’s output is the next step’s input: JSON handoff schema, exit-code contracts between steps, and how to design each stage so a failure leaves enough evidence to diagnose from the log alone — without re-running everything.
• background-jobs — Jobs that run without supervision: PID files to prevent double-runs, EXIT traps for cleanup, log rotation with date-stamped file swap. How to design failure so a crashed job leaves evidence rather than silence — and how to wire a restart policy that doesn’t loop forever.
• monitoring — Knowing a job finished cleanly without checking by hand: poll-based health checks on the output file’s last-modified timestamp, a watchdog process that re-launches on unexpected exit, and a daily summary job that reads all logs and surfaces anything that failed or stalled.
• scheduling — Cron-style task execution for Claude jobs that need to run on a clock rather than on demand.
• pipelines — Chaining Claude calls so the output of one step becomes the input of the next without manual handoffs.
• background-jobs — Jobs that run while you’re not watching — how to design them so they fail gracefully and log what matters.
• monitoring — How to know a background job is still running, finished cleanly, or silently failed without you checking it by hand.

A preview

“The goal of automation isn’t to remove yourself from the work. It’s to remove yourself from the parts that don’t need you — so the parts that do get more of your attention.”

“A pipeline that fails silently is worse than one that never ran. The difference between a useful automation and a broken one is almost always the log.”

Claude users ready to automate

Claude users who have a working daily workflow and want to stop doing the repetitive parts by hand. Prerequisites: comfortable with the Productivity daily loop.

A longer excerpt is available to newsletter subscribers.

Buy on Amazon