Type: url
Original: https://arxiv.org/pdf/2607.00038
Status: generated
Summary
A position paper that gives the mid-2026 practitioner slogan “stop prompting your agent, design the loop that prompts it” a careful treatment. It defines the loop specification — a bounded, reusable artifact of trigger, goal, verification, stopping rule and memory that a human hands to an agent harness — distinguishes it from programming loops and the harness’s internal agent cycle, gives it an anatomy and taxonomy including a five-level verification ladder, hand-codes the fifty loops of the public Loop Library, and grounds design principles and anti-patterns in the literature on self-correction, reward hacking and model-as-judge fragility. It also releases an authoring skill, sandeco-loop.