32 production-ready routines for Claude Cowork scheduled tasks and Claude Code Routines, plus a 14-pattern reliability kit. Each routine is a self-contained prompt with verification-before-report, honest fallbacks, a Desktop SKILL.md, a sample run and a failure-modes table. Includes a Weekly Reliability Review that audits your other routines and an Ops Board rebuilt from your run logs. Drafts-only by design: nothing is sent, posted, deleted or paid without you.
Why I built this: Anthropic's docs say a green routine run "does not mean the task in your prompt succeeded", and the issue tracker is full of connector calls dying with "requires approval" mid-run. I wanted routines I could actually leave alone, so every one of these verifies what it did before it reports and fails loudly when it can't. Six sample runs are public (three finish PARTIAL on purpose): https://runbook.scosovan.com/sam... . Happy to answer anything about how the guardrails work.