Gauntlet Loop

Turns an open-ended builder/critic loop into a bounded, durable, auditable, counterexample-driven improvement system. Agents may propose artifacts, diagnoses, interventions, evaluators, and actions. Durable evidence determines what happened. A write-ahead authority protocol determines what may execute. Controlled counterfactual experiments determine whether a repair works. A frozen, independent promotion kernel determines what becomes canonical. A separate release decision determines what may affect the outside world.

Use this skill when / do not use it when

Use it when a task needs durable multi-session iteration, a frozen quality spec, materialized candidates in Git worktrees, independent evaluators, counterexample memory, replayable history, non-regression promotion gates, or auditable outward effects.