원클릭으로
to-summary
Progressively summarise one large source.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Progressively summarise one large source.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | to-summary |
| description | Progressively summarise one large source. |
| disable-model-invocation | true |
Turn one source into a summary through successive reduction layers.
Require one readable source. Recommend a file or URL when asking for it; accept pasted text.
Ask once:
Any particular focus, audience, format, or length? Say “no” for a general summary.
Use Summary instructions supplied with the invocation and proceed. Complete this step when the source and Summary instructions are both clear.
Extract the source's readable text and estimate its token count.
When the source, instructions, and expected output total at most 150,000 tokens, use the full source as the final-pass input.
Otherwise, begin a reduction layer. Complete this step when every part of the source belongs to one ordered chunk of roughly 100,000 tokens or fewer. Prefer heading and paragraph boundaries.
Spawn one subagent per chunk, up to available concurrency. Give each subagent:
Collect every chunk summary in source order. Retry a missing or failed chunk. Complete the layer only when every chunk has exactly one summary.
When the ordered summaries, instructions, and expected output still exceed 150,000 tokens, group the summaries into ordered batches of roughly 100,000 tokens or fewer and run another reduction layer with the same Summary instructions and 10:1 target.
Complete reduction when the final-pass input, instructions, and expected output total at most 150,000 tokens and every original chunk reaches that final input through every layer.
Apply the Summary instructions. With none, write a broad-audience summary of at most 1,000 words in continuous prose paragraphs. Use uncited prose by default; add citations when requested.
Return the summary in chat. When the user requests a file, use their path or create a uniquely named Markdown file in the operating system's temporary directory.
Complete only when the final summary satisfies the requested focus, audience, format, and length. Deliver only the final summary.