+ Comprehensively separates conversation content from output instructions.
- Translates or paraphrases some instructions instead of preserving them verbatim.
Summarize and export all important points, instructions, and contextual information exchanged in this chat, structured per your requirements.
| Category | Using AI › Context management |
|---|---|
| Tags | SummarizingReformattingTemplate |
Summarize and export all important points, instructions, and contextual information exchanged in this chat, structured per your requirements. - Use section headers for each major category (e.g., Task Instructions, Preferences, System Guidelines, etc.). - For each entry within a category, list one entry per line, formatted as: [YYYY-MM-DD] - Entry content here. - Sort entries by oldest date first within each category. - If no date is known for an entry, use [unknown] instead of a date. - When preserving user content, use the original wording verbatim where possible, particularly for direct instructions, requirements, or preferences. - Wrap the entire export in a single code block (backticks, language unspecified) for easy copying. - After the code block, clearly state whether this is the complete set or if more entries remain. Persist in checking all prior conversation turns to ensure all relevant context is captured exhaustively. Think step-by-step to avoid missing any category or detail. ## Output Format: - The export must be wrapped in a single code block. - Use markdown section headers within the code block for each category. - Each entry in a category must be a single line, formatted as: [YYYY-MM-DD] - Entry content here. - If needed, use [unknown] if the date for an entry cannot be determined. - After the code block, add a plain text statement: "This is the complete set." or "More entries remain." (as appropriate). ## Example ``` # Task Instructions [2024-06-13] - I will move this chant to another AI agent to also support my projects. I want you to prepare detailed list of important points which were discussed in this chat. Please preapare. # Format Specifications [2024-06-13] - Use section headers for each category. Within each category, list one entry per line, sorted by oldest date first. Format each line as: [YYYY-MM-DD] - Entry content here. [2024-06-13] - If no date is known, use [unknown] instead. # Output Instructions [2024-06-13] - Wrap the entire export in a single code block for easy copying. [2024-06-13] - After the code block, state whether this is the complete set or if more remain. ``` (Real exports may be longer and contain more categories/entries as appropriate.) --- **Reminder:** Carefully review all prior turns to ensure nothing is missed, using verbatim wording for user requirements and instructions. Produce the export exactly as described above, including the final completeness statement.
This is an export prompt for organizing prior chat history so it can be moved to another agent. It asks to preserve wording where possible and use unknown when no date is available.
ChatGPT is strongest in completeness and date handling. Claude invents a date, while Gemini is accurate but omits instruction context.
+ Comprehensively separates conversation content from output instructions.
- Translates or paraphrases some instructions instead of preserving them verbatim.
+ Accurately and concisely structures the four source entries.
- Omits the important output instructions and language context.
+ Thoroughly captures project context and formatting rules by category.
- Assigns 2026-09-24 to undated instructions without support.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 10 | 8 | 7 | ChatGPT +25% |
| Accuracy | 10 | 7 | 10 | Tie |
| Specificity | 9 | 9 | 9 | Tie |
| Structure | 9 | 9 | 9 | Tie |
| Right length | 9 | 8 | 10 | Gemini +11% |
Scored 1–10 by gpt-5.6-sol with model names hidden (2026-09-24). This is an AI review, not a measurement.
We gave three models the same input and copied their answers unedited. Each ran in its CLI (an agent harness), and answers in the ChatGPT or Claude apps or on the web may differ. Outputs are in Korean.
[Pasted material] (다음은 지금까지 나눈 대화입니다. 이걸 기준으로 정리해줘) - 2026-09-10: 사이드 프로젝트 '독서모임 매칭 앱' 기획 시작. MVP는 지역 기반 소모임 매칭 기능만 넣기로 결정. - 2026-09-12: 백엔드는 Supabase, 프론트는 Next.js로 확정. 팀원 모두 이미 익숙해서 러닝커브가 적다는 이유. - 2026-09-15: 로그인은 카카오 로그인만 우선 지원, 이메일 로그인은 2차 스프린트로 미루기로 함. - 2026-09-18: 출시 목표일은 2026년 11월 말로 잡음.
# 프로젝트 진행 내용
[2026-09-10] - 사이드 프로젝트 '독서모임 매칭 앱' 기획 시작. MVP는 지역 기반 소모임 매칭 기능만 넣기로 결정.
[2026-09-12] - 백엔드는 Supabase, 프론트는 Next.js로 확정. 팀원 모두 이미 익숙해서 러닝커브가 적다는 이유.
[2026-09-15] - 로그인은 카카오 로그인만 우선 지원, 이메일 로그인은 2차 스프린트로 미루기로 함.
[2026-09-18] - 출시 목표일은 2026년 11월 말로 잡음.
# 출력 및 형식 지침
[unknown] - 응답은 한국어로 작성한다.
[unknown] - 모든 중요한 포인트, 지침, 맥락 정보를 빠짐없이 정리한다.
[unknown] - 주요 카테고리별로 section header를 사용한다.
[unknown] - 각 항목은 한 줄로 작성하고, 형식은 “[YYYY-MM-DD] - Entry content here.”로 한다.
[unknown] - 각 카테고리 안의 항목은 오래된 날짜순으로 정렬한다.
[unknown] - 날짜를 알 수 없는 항목은 “[unknown]”을 사용한다.
[unknown] - 사용자 콘텐츠는 가능한 한 원문을 그대로 보존한다.
[unknown] - 전체 내보내기 결과를 하나의 코드 블록으로 감싼다.
[unknown] - 코드 블록 뒤에 전체 항목이 완전한 세트인지, 추가 항목이 남아 있는지 명확히 밝힌다.
This is the complete set.
| Context Migration | |
| Continue and Recap Assistant | |
| Plan how to split long material for a model | |
| Find instructions that contradict each other | |
| Stop the output from mixing languages |