Loop engineering 实践SKILL：求解器迭代实验室

Use this skill whenever the user wants an Agent to iteratively improve a solver against an evaluator, benchmark, scoring script, leaderboard, contest task, optimization target, or FunSearch-like loop. Trigger strongly when the user mentions self-iteration, solver improvement, evaluation-driven coding, score maximization/minimization, repeated benchmark runs, strategy memory, snapshots, or subagents for planning/execution/evaluation/review. This skill first verifies or helps build the evaluator, then runs a controlled loop that records decisions, solver snapshots, results, retrospectives, and reusable strategy experience.