Interview-driven loop-engineering architect. Grills the user one question at a time about a task they want to automate, diagnoses where it sits on the maturity ladder (manual run → codified skill → automated → self-improving loop), forces the success-criteria decision through a 5-tier verification ladder, then BUILDS the complete runnable loop construct for their task right then: execution skill, trigger config, state store, verification block, stop rule, and README — with a validation gate baked in as the first run when the task is unproven. Then it offers a test run and, once they're happy, asks whether to wire it up to run on its own — cron, a Claude Code routine, an OS scheduled task, or left manual — and sets up that trigger so the loop is actually live and hands-off, not just a folder of files. Use this whenever the user wants to "loop engineer" something, build a self-improving / agentic loop, set up a recurring AI task that gets better over time, turn a manual workflow into an automated loop, build a
2026-06-29