--- name: "abstract_writing" description: "Draft, rewrite, and check submission abstracts for journals and conferences across paper types: structured and unstructured forms, the five-move structure of problem, purpose, method, results, and claim, word and character limit discipline with a runnable checker, vague-to-specific rewriting, and the CONSORT and PRISMA abstract reporting checklists for trials and systematic reviews. Use when an abstract must be written or fixed for a specific venue, limit, or deadline, when a draft reads vague, or when a conference abstract covers work still in progress." metadata: { "includeInPrompt": true } --- # Abstract Writing ## Purpose Produce an abstract a venue will accept and a reader can judge the work from. This skill is about submission abstracts: a named venue, a stated format, a hard limit, and often a paper type with its own reporting checklist. It covers journal and conference abstracts for empirical, review, and methods papers, plus trial and systematic review abstracts that follow published checklists. A different job, summarising a finished paper for its own front page with no venue constraints, is a plainer task. Start there only if there is no venue, no limit, and no submission at stake. ## Workflow 1. **Pin the venue spec before writing a word.** Record: word limit or character limit (and whether headings and keywords count toward it), structured or unstructured format, the exact required headings if any, paper type, and whether references, figures, or funding and registration lines are required or banned. Venue instructions override every default in this skill. Real specs vary widely: Scientific Reports caps an unstructured abstract at 200 words with no references or figures, while medical journals commonly allow 250 to 300 words in a structured format. Never draft against a guessed limit. 2. **Classify the paper type, because the moves change shape.** Empirical: the results move carries measured outcomes. Review: the method move is the search (databases, dates, eligibility rule) and the results move is counts and synthesis across included studies. Methods: the results move is benchmark performance against a named baseline under named conditions. Trial or systematic review: stop and use `references/reporting_checklists.md` first, since CONSORT for Abstracts and PRISMA for Abstracts list required items the five moves do not name, such as harms, registration, and funding. 3. **Harvest the facts before drafting.** From the paper or the author's notes, pull: the specific gap, the aim in one clause, the design in one clause (sample size, setting, data source, analysis), the primary result with its number and uncertainty, one secondary result at most, and the claim the design can support. If a fact is not in the source material, it does not go in the abstract. For work in progress, harvest only the data already analysed. 4. **Draft in five moves.** The five rhetorical moves, following the UNC Writing Center summary of abstract research: (1) problem or background with the gap, (2) purpose, what this work does about the gap, (3) method, (4) results, (5) claim or implication. Some venues compress moves 1 and 2 into one sentence. Few abstracts need more than one or two sentences per move. Write the results move first if drafting stalls: it is the part readers and reviewers weight most, and it fixes how much room the other moves get. 5. **Budget the words by move.** As a starting allocation for an unstructured empirical abstract, adapted from guidance TU Dublin attributes to George Mason University: about 25 percent purpose and importance, 25 percent method, 35 percent results, and the rest on problem framing and implications. Treat the percentages as a diagnostic, not a rule. If background takes the largest share and results the smallest, the draft is upside down and gets cut from the front. 6. **Match the form to the venue.** Structured: use the venue's headings verbatim, in its order, even when the heading names feel clumsy. Fill every heading; an empty Results heading is a rejection signal at conferences that screen for structure. Unstructured: one paragraph (some venues allow a short second paragraph for the main result), moves in the same order, no headings, and the first sentence must carry the problem rather than a general truth about the field. 7. **Handle work in progress honestly, for conference abstracts.** Analyse the data in hand and report it as preliminary with its real numbers (for example, 47 of 120 planned participants). Base every conclusion on that data only. Never write expected results, never write that results will be presented, and never promise more data collection in the abstract. Reviewer guidance is consistent on this: incomplete analysis scores lower than completed work, and invented or promised results are an integrity problem, not a style problem. If the call has a work in progress track, use it and say so in the method move. 8. **Run the checker, then read once more by eye.** Run `python3 scripts/abstract_check.py --limit --type ` and add `--format structured --headings ` for a structured venue, or `--char-limit` when the venue counts characters. The checker reports word and character counts, likely missing moves, citations or figure references, future-tense promises, and vague phrases. Move detection is a cue-phrase heuristic: treat a missing move flag as a prompt to re-read that part, not as proof. Then do the human pass the script cannot do: every number matches the paper exactly, the claim matches the design (an observational study predicts and associates, it does not show that an intervention works), and no abbreviation appears without being spelled out once unless the venue lists it as standard. 9. **Rewrite vague drafts by substitution, not by polishing.** Label each sentence with its move. Delete sentences that make no move. For each remaining vague claim, find the fact in the paper it is standing in for: a number for significant, a name for various, a count with a denominator for many studies, a baseline for state of the art. Three full before and after rewrites, for empirical, review, and methods papers, are in `references/worked_rewrites.md`. Follow that pattern and show the user the move labels when you hand back a rewrite, so they can see what changed and why. 10. **Deliver with the spec attached.** Hand back the abstract, its exact word count against the limit, the format used, and any checklist items (registration, funding, harms) you could not fill because the source material did not state them. Never silently drop a required item to fit the limit; flag it and let the author decide what to cut. ## Output Contract - One abstract, in the venue's format: exact headings for structured, a single clean paragraph for unstructured. - Word count stated as used over limit, for example 238 of 250. Character count as well when the venue counts characters. - below the abstract, a short check note: moves present, checker verdict, numbers verified against the source (or explicitly not verified), and any required checklist items missing from the source material. - For a rewrite, the abstract plus a one line reason per major change, tied to a move or a venue rule. - No citations, figure or table references, or undefined abbreviations in the abstract unless the venue explicitly requires or allows them. ## Operating Rules 1. Every factual claim in the abstract must trace to the paper or to material the author supplied. Never invent numbers, dates, sample sizes, or findings. Unverifiable means cut, never smoothed over. 2. Venue instructions beat this skill, the checklists, and habit, in that order when they conflict. Read the current call or author guidelines for the target venue before drafting. 3. Results get reported whether or not they are flattering or statistically significant. No surprises and no spin: state the primary result as it came out. 4. A claim may not outrun the design. Correlational work gets association language, a single site study gets a single site claim, a review gets claims per condition when the included studies split by condition. 5. Cut background before results, method detail before the primary result, and adjectives before numbers, in that order, when the limit bites. 6. Plain sentences, active voice where the field allows it, one idea per sentence. A sentence over about 40 words in an abstract is usually two moves crushed together; split it. 7. Style: plain academic register. No em dashes. No promotional language about the work. Words like novel and significant earn their place only when a fact next to them proves them.