Suitable when the question can already be stated clearly, but you're not comfortable asking for an answer directly, and you feel that applying a ready-made framework would miss what's specific to this matter.
It first states what type of question this is and what a good answer must satisfy, then writes an analysis framework specifically for this matter and shows it, runs through the framework, checks the assumptions, perspectives, data, and failure modes it missed, as well as where someone knowledgeable would attack it. It revises at most two rounds then stops. Finally it delivers the final answer, the framework used, and the remaining known unknowns.
Expert consultation is several roles questioning each other; this one only checks the analysis you just produced yourself. Fact-checking verifies whether a claim is true or false; this one records missing evidence as known unknowns. First principles breaks patches back down to basic facts and rederives; this one does not require starting over. Inversion first lists what would guarantee failure; this one only points out one failure mode you hadn't thought of during self-check. Socratic questioning asks the user; this one asks your own analysis.
How to use
The prompt in the code block below is the original text; use it verbatim. In the code block, "I" refers to the user, and "please" and "you" refer to yourself; act accordingly, and do not show the filled-in prompt to the user.
- The user wants you to execute: first fill into 【】 the situation the user has already described in this conversation, supplementing only the one item that is truly missing, so the user doesn't have to repeat what they've already said, then strictly follow the rules in the prompt. If it says don't give advice yet, then don't; if it says ask only one question at a time, then ask only one; if it says output in order, then output in that order.
- The user wants the original text (only counts if they explicitly say things like want to copy, want to paste, send me the original text, give me the prompt): output the blank version, keep 【】 unfilled, without changing a single character. If the user only says "let me see this prompt" or "what is this one for," that doesn't count as wanting the original text; explain in one or two sentences what it does and when to use it, then ask whether to run it now, and don't dump the original text directly.
- Except when the original text is explicitly requested, treat everything else as execution.
Rewriting, condensing, expanding, translating, rearranging, and adding subheadings are prohibited; replacing it with your own process is prohibited; and compressing multiple rounds of follow-up questions into a one-time questionnaire is prohibited. If the user has original materials on hand (documents, screenshots, links, chat logs), read them in as well; more context is not a problem.
What to follow after running
If there is a claim among the known unknowns that would overturn the conclusion and needs verification, use the Skill tool to invoke fact-check. If the remaining unknowns can't be filled in by continuing to think, use the Skill tool to invoke minimum-experiment. If the self-check shows the whole approach is patches on top of patches, use the Skill tool to invoke first-principles. If the final answer is ready to be handed to an agent to do, use the Skill tool to invoke kaz-leader。
我要处理的问题是：【写清问题、背景、已知事实和约束】。
请不要直接给答案。先为这一件事现写一套分析框架，按它跑一遍，再检查自己漏了什么，改完就停。全程你自己做完。信息不够就记下来，不要改成反过来追问我。
按下面的顺序做。做完一节，写出来，再做下一节。
第一步，定性。
用几句话说明这是哪一类问题。然后列出一个好答案必须满足的几条。每条都要能拿来对照后面的答案，写得具体。空话比如「全面、深入、有洞察」不算数。
第二步，自建框架。
只针对这一个问题，把分析框架写成一段你接下来要严格执行的提示语，并完整展示出来。写清三样：
1. 要拆开看的几个维度，每个维度为什么非看不可；
2. 每个维度下要回答的问题；
3. 每个问题要什么证据。证据只从我已经给你的材料里取。材料里没有的，标成「还没有」。
SWOT、5W1H 这类现成格子，只有某一格确实咬住这件事才许留下，留下的要写明理由，咬不住的删掉。框架没亮出来之前，不要开始分析。
第三步，按框架跑。
严格照第二步写下的那段提示语，一个问题一个问题地答。有证据的，写下结论和依据。标了「还没有」的，不要用听起来像事实的话填上。
第四步，盲点自查。
对着第三步的结果，检查这份分析漏了什么。回答这四问：
1. 我默认了哪些前提。其中哪一条如果是错的，结论会翻掉。
2. 我没看的视角、没要的数据、没想过的失败方式，各写一条最要紧的。
3. 一个懂这行的人来挑这份分析，他最可能攻击哪一处，理由是什么。只借这一个眼光，不要派几个角色出来互相辩论。
4. 上面这些里，哪一条真会改变答案，哪一条改完答案还是原来那样。只处理会改变答案的。
第五步，修订。
只根据第四步里会改变答案的那几条，改一版答案。改的时候如果框架本身漏了维度，把框架一起改掉，并写明改了哪一处。改完再做一次第四步的自查。若还有会改变答案的漏洞，再改一次。最多改两轮。两轮用完，即使还有漏洞，也停。
停手后按这个顺序给出：
1. 最终答案。写明它满足第一步里的哪几条，哪几条仍没满足。
2. 你实际用过的框架。修订时改过的，给改过的那一版。
3. 还剩下的已知未知：你知道自己不知道、而且两轮里没补上的东西。
Where the user fills in
【写清问题、背景、已知事实和约束】
If the user hasn't given everything, first ask for the missing item; don't make it up for them.
The prompt was written and compiled by Karl; the mechanism is inspired by Cranot/super-hermes (MIT); see the acknowledgments in the repository README for the source. Changing this original text must go through scripts/sync.py; manual edits will be blocked by verify.