Set up, scaffold, and administer agentic loops in Claude Code — headless while-loops around `claude -p`, evaluator-optimizer (generator/critic) loops, meta / prompt-refinement loops (a loop that refines the prompts another loop runs), and orchestrator fan-out. Every loop ships with non-negotiable guardrails: a verifiable exit condition, a max-iteration cap, and in-code budget enforcement. Use whenever the user wants to build, design, run, wire, or supervise loops / agent loops / "loop engineering" / nested sub-loops, or asks how to make Claude prompt Claude on a schedule or until a goal is met.
2026-06-30