菜单
Use at study design and again before writing, to check that every deliverable the task statement names has been produced. Covers how to enumerate what was asked for, why partial coverage scores worse than it feels, and what to do when a named deliverable is out of reach.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/tangxiangru/AutoR --skill cover-what-the-task-named命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | cover-what-the-task-named |
| description | Use at study design and again before writing, to check that every deliverable the task statement names has been produced. Covers how to enumerate what was asked for, why partial coverage scores worse than it feels, and what to do when a named deliverable is out of reach. |
Read the task statement and list, literally, every output it names: each model to be built, each dataset to be evaluated on, each baseline to compare against, each quantity to report. Write that list down at design time, before you have results, and carry it to the end.
This matters more than it looks. A study that does one arm of the task thoroughly and skips the other two does not score two-thirds — evaluation of research asks, per requirement, whether the work is there, and a requirement with nothing behind it scores zero however good the rest is. Depth on one arm does not pay for absence on another.
The common shape is: the task names three experiments; the run finds the first one interesting; the report is an excellent study of the first one and does not mention the other two. From the inside this feels like focus. From the outside two thirds of the work is missing.
The second shape is subtler: the task names a comparison ("against the incumbent method", "across both datasets") and the run reports its own numbers without the comparison. A number with nothing to compare to is not a result.
Some named deliverables genuinely cannot be produced — the data is not supplied, the compute is not there, the reference implementation is not public. That is a real answer, and it is worth writing down properly: name the deliverable, say exactly what is missing, and say what you did instead. One paragraph.
What is not acceptable is silence. A deliverable that is neither produced nor mentioned reads as one the run forgot.
Go back to the list. For each entry: produced, or explained. Nothing unmarked.
