auto-research-skills

Structures long-horizon, open-ended, research-heavy, and optimization-driven tasks as constrained iterative loops with scorecards, experiment logs, and keep-or-revert decisions. Use when progress should come from repeated exploration, measurement, and frontier management rather than a one-shot answer.

将长周期、开放式、研究型和优化型任务组织为带记分卡、实验日志和保留或回退决策的约束式迭代循环。适用于需要依靠持续探索、度量和 frontier 管理推进，而不适合一次性完成的任务。