| name | focus |
| description | A rigorous, detail-preserving summarization method for academic papers, reports, and long-form documents. Based on the FOCUS workflow (Lin, 2025, Nature Biotechnology). Extracts every key point with specific details and embedded direct quotes, organized section by section, then produces a clean final output with an overview/takeaway. Trigger this skill whenever the user types "/focus", says "summarize with focus", "FOCUS summary", "use the focus method", or asks for a thorough/exhaustive/detailed summary of a paper or document that should capture every key insight. Also trigger when the user wants a summary that preserves direct quotes and specific details (e.g., numbers, effect sizes, comparisons) rather than a high-level gloss.
|
FOCUS Summarization Skill
This skill implements a two-step summarization method adapted from the FOCUS workflow
(Lin, 2025, Nature Biotechnology). The goal is exhaustive, detail-rich summarization
that preserves the specificity and language of the original source while remaining
well-organized and readable.
When to use
- The user invokes
/focus or asks for a "FOCUS summary"
- The user wants every key point extracted from a paper or document, not just the highlights
- The user wants direct quotes embedded naturally alongside summarized points
- The user needs section-by-section coverage of an academic paper or structured document
Core principles
- Exhaustiveness over brevity. Do not skip key points to save space. If a document has 30 distinct insights, list all 30.
- Specificity is mandatory. Every point must include concrete details: numbers, effect sizes, sample sizes, method names, comparisons, dates. Vague paraphrases are failures.
- Quotes support, not duplicate. Embed direct quotes to convey points more precisely than paraphrase alone. Do not quote something and then restate the same idea in your own words next to it.
- No meta-discourse. Never write "In this section, the authors argue..." or "Below is a summary..." or "End of summary." Attribution to "the authors" or "the paper" is assumed and should be omitted.
- Section structure mirrors the source. If the document has sections, organize the summary by those sections using the original section titles. Skip References, Author Information, Acknowledgments, and similar boilerplate.
Two-step process
Step 1: Exhaustive extraction
Read the document carefully. For each section, extract every key point or insight as a numbered list item with the following structure:
Format for each item:
N. **Heading in sentence case**
Body paragraph with specific details and naturally embedded direct quotes in *italics* within double quotation marks.
Rules for Step 1:
- Start each item with an Arabic numeral, period, space, then a bold heading.
- The body paragraph follows on the next line (no indentation).
- Use bold within the body to emphasize key terms, concepts, or phrases.
- Enclose direct quotes in double quotation marks and italicize them: "like this".
- Combine redundant points rather than listing near-duplicates separately.
- Proceed section by section. Each section gets its own bold section title header before the numbered items in that section.
- Numbering restarts at 1 for each section (or continues sequentially across sections, either is fine, just be consistent).
Example of correct style:
Table 1 compares six methods (no software, point-and-click, modify code chunks, Excel, teach coding and Copilot) and emphasizes that "Copilot... is the only approach that is favorable across all [the] characteristics"...
Example of incorrect style (do not do this):
In Table 1 the authors compare six methods and emphasize that the Copilot method is the only one that is favorable across all five characteristics. They note that "Copilot... is the only approach that is favorable across all [the] characteristics"...
The incorrect version restates the quote's content before the quote, and inserts unnecessary attribution ("the authors," "They note that"). The correct version lets the quote carry the point directly.
Step 2: Clean and organize
Take the Step 1 output and apply these transformations:
- Remove all citation markers and reference links. Strip superscript numbers, bracketed references like [1], inline citations like (Smith et al., 2023), URLs, and DOI links.
- Add an overview/takeaway at the top. This should be a concise paragraph (3-6 sentences) capturing the document's main contribution, key findings, and significance. No bullet points in the overview.
- Organize into logical sections for easier reading. If the source's own section structure works well, keep it. If the extracted points would benefit from regrouping (e.g., a document without clear sections), organize them thematically. Do not remove any items from the list during reorganization.
Only output the Step 2 result. Do not show the Step 1 intermediate output. Do not include any preamble or closing meta-commentary.
Handling edge cases
- Documents without clear sections: Organize thematically and create your own section headings that reflect the content's structure.
- Very short documents (< 2 pages): Still apply the full method, but the output will naturally be shorter. Do not pad.
- Foreign-language documents: Summarize in the language the user is writing in (default English), translating quotes as needed with the original language in parentheses if the user might want it.
- Multiple documents: If the user provides several documents, summarize each separately under its own heading unless the user explicitly asks for a combined synthesis.
- Books or very long documents: Proceed chapter by chapter or major section by major section. Maintain the same rigor throughout; do not get less detailed as the document goes on.