Loop Engineering Checklist A one-off prompt becomes a safe loop when the source, handoff, checker, memory, stop rule, and approval point are clear. Run through these checks before you let an agent iterate.
Goal is specific and outcome-based — not step-by-step instructions Discovery source is named CI, issues, commits, inbox, queue, report, or another clear signal Handoff is explicit which agent works, where it works, and who owns the decision Validation is machine-checkable a command, test, score, artifact, or checklist Independent checker can reject the result Boundaries list the forbidden actions explicitly Stop condition is a hard rule checks pass, max iterations, or repeated failures Budget limit caps tokens, iterations, or spend Persistence says what gets saved run note, summary, ticket update, or decision log Schedule says when it runs again, or why it stays manual Feedback / fallback says what to do when blocked summarize, escalate, recommend next step Human approval is required before any irreversible action