Create Abstract

/create-abstract

Produce a publication-quality abstract as a LaTeX fragment, built on the six-part model and the house style rules. The researcher is expected to adapt the result; your job is to hand them text that would already pass an editor.

Before anything else

1. Load the style rules. Invoke the paper-writing-style skill and read its references/human-voice.md and references/latex-conventions.md. If that skill is not installed, fall back to the condensed rules at the bottom of this file.
2. Lock the language. Detect the language of the user's first message and keep using it for every interaction in this run. Never switch mid-run, even if the topic is English prose. The LaTeX output is always English, whatever the interview language.
3. Read the context. Look for .tex files, a .bib file, and any existing abstract in the working directory. Detect the citation style as described in the LaTeX conventions. Do not ask the user for anything you can find yourself.

The six-part model

Purpose, results and conclusions are mandatory. Without them the text is not an abstract. Contextualization and gap are strongly recommended, and may be dropped or merged under a tight word limit or a specialist audience. Two parts can share one sentence when space is short.

Total is normally 200--300 words unless the journal says otherwise.

Workflow

1. Gather what exists

Accept any of: a short description of the work, paths to a manuscript or a results section, a figure list, or nothing at all. Read whatever is given before asking a single question.

2. Interview

Ask in batches of at most four questions, using AskUserQuestion where the answers are selectable. Keep going until you can write every mandatory part without inventing anything. Cover:

• Target journal. This governs everything downstream: word limit, whether the journal wants a structured abstract with labelled sections, and how much contextualization to spend. A specialist journal needs little field justification because the audience already has it; a broad journal such as Nature or Science needs the importance of the whole field established. If the researcher does not know yet, use 200--300 words, the full six-part model, and a specialist audience.
• The broad field, in the terms the target audience uses.
• The gap. Press on this. A weak gap is the most common reason an abstract reads as unremarkable. What is unresolved, unexplained, contested, or too expensive in the current state of the art? The purpose must answer this specific gap.
• The purpose, and how it addresses that gap.
• Methods, at the level of technique names rather than parameters.
• The principal result, with actual numbers. Not all results, the ones that matter. Ask for the values; never estimate them.
• The contribution: what a reader in the field can now do, or now knows, that they could not before.

Do not draft before the mandatory parts are covered. If a number or citation is genuinely unavailable, that becomes a % TODO:, not a guess.

3. Draft

Emit one LaTeX fragment, in English, labelled with structure comments so the researcher can see the model being followed:

% CONTEXTUALIZATION
Silver nanoparticles are widely used as antimicrobial agents in wound dressings
\citep{silva2024}.
% GAP
However, their long-term cytotoxicity towards human keratinocytes remains poorly understood.
% PURPOSE
In this paper, we report the cytotoxicity of citrate-capped silver nanoparticles towards
HaCaT keratinocytes over 72\,h.
% METHODS
Cell viability was assessed by MTT assay and membrane integrity by lactate dehydrogenase
release.
% RESULTS
Viability fell to 62\% at 25\,\textmu g\,mL$^{-1}$ and to 18\% at 100\,\textmu g\,mL$^{-1}$,
while lactate dehydrogenase release increased threefold over the same range.
% CONCLUSIONS
These results indicate a concentration threshold below which silver dressings remain
compatible with keratinocyte monolayers, which is directly relevant to the design of
antimicrobial wound care materials.

Show it in chat in a fenced latex block. Then state the word count and, in two or three lines, which choices you made and why, so the researcher can argue with them.

4. Check and refine

Ask what to change. Expect pushback on the gap and on which result leads; those are judgement calls only the researcher can settle. Iterate as many rounds as they want, showing the full fragment each time rather than a diff, since the abstract is short and reads as a whole.

5. Offer to write it out

Only once the researcher is satisfied, offer to write the fragment to a file or patch it into the manuscript. Never write to their files before that point.

Hard rules

• Never invent a result, a number, a method detail, or a citation key. Emit % TODO: <exactly what is missing> instead.
• Never emit an em dash (---). -- only in numeric ranges. Use commas, colons, semicolons, and full stops.
• Past tense for what was done, present tense for what the results indicate.
• Third person by default. First person plural is fine sparingly, for emphasis.
• The main finding appears as early as the model allows. There is no suspense in a scientific abstract.
• Cut it is important to note that, plays a key role, novel, successfully, carefully, and every sentence that would fit unchanged in a different paper.

Condensed style rules, if paper-writing-style is not installed

Clarity and concision govern. One idea per sentence, short enough that the reader reaches the full stop with the opening still in mind. Be specific: a sentence that could appear in any paper in the field says nothing. Put the action in the verb (the condition improved, not an improvement in the condition was observed). Cut redundancy (completely eliminated, introduce a new). Prefer plain words (use over utilization, to over in order to, most over the majority of, because over due to the fact that). Avoid delve, harness, leverage, underscore, pave the way, crucial role. Keep however, moreover, in contrast, in this paper, here we report: those are correct scientific register, not filler. Never stack passive voice. Make sure every it and this has one possible referent. Thin space between quantity and unit (10\,nm), escape \%.