long-running-agent-tasks

A meta-skill for designing and generating "long-running task Skills". Use when a user needs to create a Skill that involves a large number of files, requires cross-session resumption, or needs concurrent subagent scheduling. Trigger scenarios: (1) the user says "help me make a skill that bulk-processes XXX", (2) the user describes a batch task involving dozens to hundreds of files, even if they don't mention "skill", (3) the user says "create a long-running task", "make a long-running task skill", "a skill for bulk migration/review/fix". This skill drives the entire generation process end-to-end, calls skill-creator's eval capability in Step 5 for a smoke-test (user confirmation gate), and uses an independent grader agent in Step 6 for artifact quality review.

Skill 执行质量单次评测工具。对已有 Skill 进行一次端到端质量评估：生成测试用例、运行 Skill 产出、 通过 Grader 对 assertions 评分、聚合 benchmark 统计、启动可视化评审页面呈现结果。 当用户说"帮我评测这个 skill"、"跑一下测试看看效果"、"对比一下有没有 skill 的区别"、"run evals"、 "benchmark this skill"、"评估 skill 质量"时使用。只做单次评测出报告，不含迭代改进循环。