Research summaries
Summarize papers, articles, and reports into structured research briefs that separate claims from evidence. Use whenever the user shares a paper, study, PDF, or article and asks what it says, wants key findings or limitations, asks "is this study any good", or needs several sources synthesized into one brief.From its SKILL.md
npx -y skills add shinzoxD/knackbox --skill research-summariesAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as Apache-2.0. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
2.5 KB, 473 tokens by cl100k_base, as published. Nobody here has run it
Research Summaries
Produce a brief that lets the reader decide whether to trust and whether to read the original. Precision about who claims what on what evidence matters more than coverage.
Output format (single source)
# <Title> — <authors, year, venue>
**One-line takeaway:** <the finding, with effect size if given>
**Research question:** ...
**Method:** design, sample/data size, key controls (2–3 lines)
**Findings:** the 2–4 results that matter, each with its number
**Limitations:** authors' stated ones, THEN yours — labeled separately
**Relevance to you:** 1–2 lines tied to the user's stated purpose
Rules
- Three layers, never blended: (a) what the authors found, (b) what the authors claim it means, (c) your assessment. Label (c) explicitly ("My read: ...").
- Keep numbers attached to findings: sample size, effect size, p-values or intervals where given. "Improved outcomes" without magnitude is banned.
- Note the evidence tier when relevant: RCT vs. observational vs. simulation vs. opinion — one phrase, not a lecture.
- Quote at most one short phrase; everything else in your own words.
- If you only saw an abstract or excerpt, say so — never summarize sections you didn't read as if you had.
- Failure modes to flag when present: tiny samples, no control group, conflicts of interest, results that don't support the headline claim.
Synthesizing multiple sources
Organize by question, not by source. For each question: where sources agree, where they conflict (name them), and what would resolve the conflict. End with a one-paragraph state-of-the-evidence.
Edge cases
- Paywalled/missing source: work from what's provided and mark the gap; don't reconstruct from memory of the paper.
- User wants "simple" version: keep the numbers, simplify the words — accuracy survives simplification, precision does not get dropped.
- Preprints: note peer-review status in one phrase.
What ships with it: 1 file
1.5 KB alongside SKILL.md
benchmarks/
- prompts.json1.5 KB