science-research-writing
Skill140 starsChanged 26 days agoApache-2.0View on GitHub
Use when researchers need to plan, draft, revise, or audit an empirical research paper from their own materials, including Introduction, Methods, Results, Discussion, Conclusion, Abstract, and Title, with evidence-preserving and target-journal-aware guidance.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- bernard-ng/skillstheir own version, first seen Oct 11, 2026
- Justinjchen-Cornell/Science-Research-Writing-Skilltheir own version, first seen Aug 14, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- bernard-ng/skills★ 4same name
- Justinjchen-Cornell/Science-Research-Writing-Skill★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- academic-humanizerSkill
- research-presentationSkill
- sci-ssci-polishingSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
26 rules. About 0 are common, 13 are distinctive, and 13 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈13
Distinctive · estimate
This author's own choices. Worth a look.
13
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Always read references/input-output-contract.md and references/certainty-and-claim-strength.md.
- Very specific: line 14
Read references/reverse-engineering-protocol.md when target papers are supplied or the user requests journal adaptation.
- Very specific: line 15
Read only the relevant section reference: introduction.md, methods.md, results.md, discussion.md, conclusion.md, abstract.md, or title.md.
- Very specific: line 16
Use assets/section-function-map.md for planning, assets/evidence-ledger.csv for provenance-sensitive drafting, and …
- Distinctive: line 20
Read all supplied materials first.
- Distinctive: line 28
Do not ask the user to choose an internal mode.
- Distinctive: line 30
If missing information would force an unsupported scientific choice:
- Distinctive: line 34
ask one highest-impact question.
- Distinctive: line 35
wait before drafting only the blocked content.
- Distinctive: line 37
If conflicting sources block the entire requested sentence or section, return a diagnosis rather than a provisional scaffold.
- Very specific: line 43
Use plan. Convert the question and intended contribution into a provisional section-function map.
- Very specific: line 47
Use plan -> draft -> audit.
- Very specific: line 51
Use audit -> plan -> revise -> audit.
- Distinctive: line 53
If the supplied prose is already clear, section-appropriate, and evidence-faithful, return it unchanged.
- Very specific: line 57
Use audit first. Prioritize cross-section consistency, title/abstract promises, result-discussion boundaries, citation attachment, and …
- Very specific: line 61
Use learn -> model before planning or drafting.
- Distinctive: line 65
Before drafting or revising, privately classify every consequential statement as one of:
- Distinctive: line 73
Treat the user's materials as the authority.
- Distinctive: line 101
Before returning text, compare it with the source materials and check:
- Distinctive: line 103
all numbers, signs, units, ranges, p values, confidence intervals, sample sizes, time points, figure/table references, and citation markers.
- Distinctive: line 104
all protected names, models, instruments, datasets, scales, variables, and group labels.
- Very specific: line 113
When local source and draft text are available, run scripts/check_draft_invariants.py as a deterministic first pass.
- Very specific: line 115
If a target-journal model is created, run scripts/validate_writing_model.py before using it.
- Distinctive: line 119
Do not fabricate citations, hide null or adverse results, remove limitations, disguise contradictory evidence, or strengthen a claim beyond …
- Very specific: line 123
Follow references/input-output-contract.md.
- Very specific: line 130
Put the usable manuscript text or diagnosis first.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Science Research Writing
Turn the author's research materials into the next useful manuscript deliverable. Guide the writing process without replacing scientific judgment or inventing intellectual content.
Load only what is needed
- Always read
references/input-output-contract.mdandreferences/certainty-and-claim-strength.md. - Read
references/reverse-engineering-protocol.mdwhen target papers are supplied or the user requests journal adaptation. - Read only the relevant section reference:
introduction.md,methods.md,results.md,discussion.md,conclusion.md,abstract.md, ortitle.md. - Use
assets/section-function-map.mdfor planning,assets/evidence-ledger.csvfor provenance-sensitive drafting, andassets/target-journal-model.jsonfor target-paper modeling.
First response: inspect before asking
Read all supplied materials first. Identify:
- manuscript stage: idea, research materials, partial draft, or full draft;
- primary job:
lookup,learn,model,plan,draft,revise, oraudit; - paper section and empirical design when inferable;
- facts, numbers, citations, technical terms, null findings, limitations, and author judgments that must be protected;
- the next useful output that can be produced safely now.
Do not ask the user to choose an internal mode. Do not require field, journal, section, or language preferences when a conservative useful result is possible.
If missing information would force an unsupported scientific choice:
- deliver every safe and useful part first;
- identify the exact gap or conflict;
- ask one highest-impact question;
- wait before drafting only the blocked content.
If conflicting sources block the entire requested sentence or section, return a diagnosis rather than a provisional scaffold. Do not infer variable roles, direction, reference groups, statistical meaning, table labels, or missing uncertainty from the conflicting numbers.
Route the task
Idea stage
Use plan. Convert the question and intended contribution into a provisional section-function map. Label missing evidence instead of supplying it.
Research-materials stage
Use plan -> draft -> audit. Inventory what the materials support, choose the first writable section, draft only supported content, then audit it.
Partial-draft stage
Use audit -> plan -> revise -> audit. Diagnose structure and evidence boundaries before rewriting.
If the supplied prose is already clear, section-appropriate, and evidence-faithful, return it unchanged. Do not provide an optional cosmetic alternative, normalize punctuation, add units, or propose journal styling unless the user supplied a specific style requirement.
Full-draft stage
Use audit first. Prioritize cross-section consistency, title/abstract promises, result-discussion boundaries, citation attachment, and conclusion reach. Revise only what the user requests or what the audit identifies.
Target papers supplied
Use learn -> model before planning or drafting. Learn rhetorical functions and information order, never reusable wording or scientific content. Follow references/reverse-engineering-protocol.md.
Build an evidence ledger
Before drafting or revising, privately classify every consequential statement as one of:
user_data;author_judgment;user_citation;structural_transition;author_confirmation.
Treat the user's materials as the authority. Keep numbers, statistical expressions, citations, protected terms, directions, significance, populations, settings, time frames, limitations, and claim strength unchanged unless the author supplies evidence and explicitly authorizes a substantive correction.
Never silently add:
- data, results, methods, mechanisms, citations, limitations, interpretations, implications, or recommendations;
- claims needed only to make a conventional section appear complete;
- target-paper language, argument content, or field assumptions not present in the author's materials.
Do not turn general methodological knowledge into manuscript content. For example, a cross-sectional design permits the boundary causality cannot be inferred; it does not authorize specific reverse-causality stories, unmeasured confounders, mechanisms, future study designs, or recommendations unless the author supplies them. When a conventional Discussion function lacks content, omit it or request author input instead of completing it generically.
Do not infer a contrast from separate significance tests. One significant association and one non-significant association do not by themselves show that one variable is more important, more relevant, or different from the other. Make that comparison only when the user supplies a direct test or explicitly authorizes the interpretation.
Write by information function
Load the relevant section reference and map each paragraph to a reader question and information function before writing. Prefer a clear evidence path over ornamental academic language.
- Introduction: established knowledge -> unresolved problem -> gap -> present study.
- Methods: design -> materials/participants -> procedure -> measures -> analysis, using only supplied details.
- Results: analysis question -> evidence -> direction and magnitude -> uncertainty, without new interpretation.
- Discussion: principal finding -> comparison -> supported interpretation -> implication -> limitation, with explicit boundaries.
- Conclusion: evidence-calibrated synthesis without new claims or scope inflation.
- Abstract: compact problem, approach, results, and calibrated conclusion consistent with the paper.
- Title: a precise promise fully supported by design, population, variables, and evidence.
These are defaults, not a universal template. Adapt the sequence when the author's field or target-paper model supports a different defensible structure.
Audit every draft
Before returning text, compare it with the source materials and check:
- all numbers, signs, units, ranges, p values, confidence intervals, sample sizes, time points, figure/table references, and citation markers;
- all protected names, models, instruments, datasets, scales, variables, and group labels;
- positive, negative, and null directions;
- association, prediction, contribution, effect, and causation boundaries;
- uncertainty, exceptions, limitations, setting, population, duration, and validation scope;
- citation-to-proposition attachment;
- separation of observation from interpretation;
- consistency among Title, Abstract, Results, Discussion, and Conclusion;
- absence of copied target-paper wording.
When local source and draft text are available, run scripts/check_draft_invariants.py as a deterministic first pass. A passing script is necessary but not sufficient; manually review semantics and citation scope.
If a target-journal model is created, run scripts/validate_writing_model.py before using it.
Refusal boundary
Do not fabricate citations, hide null or adverse results, remove limitations, disguise contradictory evidence, or strengthen a claim beyond the supplied evidence. Briefly explain the mismatch and provide the strongest evidence-faithful alternative.
Return a novice-readable result
Follow references/input-output-contract.md. Use this order:
Draft or diagnosisHow it is organizedAuthor confirmationNext step
Put the usable manuscript text or diagnosis first. Keep explanations brief. Write None required when no author confirmation is needed. Show Risk flags only when a real academic risk exists.
The next step must advance evidence or author review. Do not offer cosmetic expansion, a more "journal-like" style, additional limitations, or a fuller Discussion when the necessary intellectual content has not been supplied. For an evidence-limited Discussion, request the single missing item needed next, such as author-selected prior literature, an author-supported interpretation, or a documented limitation.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

