Science Writing

Science Writing

Use this skill to build manuscript argument before sentence polish. Science-family writing must make a broad reader care, then make a specialist trust the evidence.

Core stance

• Author evidence comes first. Do not invent results, sample sizes, mechanisms, novelty,

statistics, references, datasets, code, ethics approvals, or limitations.

• Make the advance legible to a reader outside the narrow subfield.
• Use broad significance only when it is anchored to specific evidence.
• Prefer decisive, bounded claims over inflated novelty.
• If a required input is missing, draft a scaffold with placeholders and flag the gap.

Open extra files

| File | Open when | |---|---| | [references/article-architecture.md](references/article-architecture.md) | Need section-level argument order, article-type routing, or full-outline structure | | [references/abstract-introduction-results.md](references/abstract-introduction-results.md) | Drafting or revising abstracts, introductions, Results openings, or Discussion arcs | | [references/science-corpus-observations.md](references/science-corpus-observations.md) | Need article-sample observations, title patterns, or Science-family style heuristics | | [references/chinese-author-workflow.md](references/chinese-author-workflow.md) | User provides Chinese notes, mixed Chinese-English lab notes, or asks for Chinese explanation |

Intake

Before drafting, identify:

• target journal and article type, if known
• paper type: discovery, mechanism, method, resource, clinical, translational, robotics,

materials, computational, policy, or interdisciplinary

• central advance: what is newly shown, built, measured, or explained
• evidence: figures, tables, controls, comparisons, datasets, experiments, statistics
• audience bridge: why a broad Science reader should care
• boundary: where the claim stops

If central advance, evidence, or boundary is absent, expose the gap before writing.

Workflow

1. Write a one-sentence argument:

In [system/problem], we show [advance] using [approach], supported by [evidence], with [boundary].

1. Choose a section architecture from references/article-architecture.md.
2. Build a claim-evidence map before prose:

Claim | Evidence | Figure/table/source | Boundary | Confidence.

1. Draft from the evidence outward. Keep the strongest result near the central claim.
2. Calibrate verbs: show, demonstrate, reveal, enable, suggest, indicate,

may, could.

1. Remove unsupported "first", "novel", "universal", "transformative", and "paradigm"

claims unless the user provided evidence.

1. Check that the first paragraph answers: relevance, gap, and why now.
2. Return the draft plus assumptions, missing inputs, and the claim-evidence map.

Section defaults

Title

Prefer concrete titles:

object/system + mechanism/capability + consequence or setting

Good title jobs:

• name the scientific object
• signal the advance
• avoid grant-style aims and slogans
• remain searchable

Abstract

Default pattern:

broad context -> unresolved gap -> approach -> decisive result -> implication -> boundary

For methods, AI, and robotics:

task bottleneck -> design principle -> system/method -> fair comparison -> capability -> boundary

For translational studies:

clinical problem -> biological or technical gap -> model/cohort/intervention -> primary finding -> translational meaning -> limitation

Introduction

Use an hourglass:

field-scale relevance -> specific bottleneck -> what prior work cannot resolve -> this study's move -> preview of evidence

Do not turn the final paragraph into a mini-Results section. State what the paper does and why that resolves the gap.

Results

Use an evidence ladder:

system/workflow -> validation -> main result -> comparison/control -> mechanism/diagnostic analysis -> generalization/application

Each subsection should open with a claim and then prove it.

Discussion

Use:

central advance -> interpretation -> relation to prior work -> constraints -> future use

Do not re-list every figure. Translate the evidence into meaning and boundary.

Output format

Default:

1. Draft: with the requested prose.
2. Claim-evidence map: for major claims.
3. Structure notes: with the section logic.
4. Assumptions / missing inputs: with only material issues.

For Chinese author notes, provide polished English first, then brief Chinese notes explaining major structural choices and unresolved author decisions.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: shikaijieskj
• Source: shikaijieskj/science-skills
• License: MIT

Install and usage instructions live in the source repository linked above.