长文分层摘要Layered long-document summary

把长文摘要成层级要点并保留出处。 输入：长文原文、章节结构、摘要长度上限。

Output an overview, section points and open questions; each point carries a paragraph locator. Inputs: Long source text, section structure, and summary length limit.

总结原文。输出 overview、sections[]（heading、points[]、locator）、open_questions[]。每个 point 必须来自对应段落，locator 写章节/段号。不得把作者未来工作写成已完成结论。总长度不超过 {limit}。

占位符用 {字段} 标注，运行前替换为本次输入。

Summarise the source. Output overview, sections[] (heading, points[], locator), and open_questions[]. Each point must come from its paragraph; locator gives section/paragraph number. Do not turn future work into completed conclusions. Keep total length under {limit}.

Replace {field} placeholders with the current inputs before running.

overview、sections[]、open_questions[]。

overview, sections[], and open_questions[].

• 每条 point 有 locator 且可回连。
• 未来工作仍在 open_questions。
• 长度不超上限。

• Each point has a locator and can be traced.
• Future work remains in open_questions.
• Length stays under the limit.

• 摘要混入观点：删除评价句。
• 定位缺失：补段号或重分节。

• Opinions enter the summary: remove evaluative sentences.
• Missing locators: add paragraph numbers or re-section.

提示词效果未实测。机制依据见出处；换模型、版本或输入会改变结果。

This site has not tested the effect of this prompt. See the sources for the mechanism; changing model, version or input can change the result.