# Nature Polishing

> Polish, restructure, translate, or humanize academic prose into Nature-leaning English using writing-strategy principles, curated Nature/Nature Communications article patterns, and phrase-level support from Academic Phrasebank. Use whenever the user asks to polish a manuscript paragraph, abstract, introduction, results, discussion, conclusion, title, methods section, or Chinese academic draft for publication-quality English, or when the user wants an already polished paper to sound less mechanical through a controlled final human-style pass.

- Skill: `lingwei-zheng/nature-polishing` (Agent Skill, multi-file: 11 files)
- Install (CLI): `npx skillmds@latest add lingwei-zheng/nature-polishing`
- Raw SKILL.md: https://api.skillmd.com/api/skills/lingwei-zheng/nature-polishing/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: lingwei-zheng (https://skillmd.com/u/lingwei-zheng)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/lingwei-zheng/nature-polishing

---


# Nature-Style Academic Polishing

Improve scientific prose without changing its verified scientific content. Treat
`Nature-leaning` as concise, evidence-led, readable academic English, not as an
instruction to imitate a journal mechanically. Follow the user's target journal,
house style, and explicit preferences when they differ from the defaults here.

## Modes

Choose the smallest mode that matches the request.

- `formal-polish`: default mode for translation, restructuring, section rewriting,
  clarity, concision, academic register, and publication-quality English.
- `human-final-pass`: use only after formal polishing when the user wants credible
  academic prose that sounds less mechanical.

For `human-final-pass`:

- Mention once that it is intended for already polished text, then continue.
- Preserve facts, citations, formulas, LaTeX, numbers, technical terms, and claims.
- Apply natural variation only to eligible prose in Introduction, related-work
  Literature Review, Results or result-facing Discussion, Conclusion, and
  Limitations.
- Do not humanize Methods, formulas, tables, references, code, exact technical
  claims, or numeric statements.
- Never add grammar errors unless the user explicitly enables grammar-error mode
  in the same request.
- Read the three `human-final-pass-*` references before making substantial changes.

## Non-Negotiable Boundaries

- Repair argument and section logic before sentence style.
- Do not invent data, references, mechanisms, evidence, or novelty claims.
- Do not upgrade association to causation or broaden generalisability.
- Treat an accurate associational verb as a sufficient causal boundary in
  ordinary prose. Do not automatically add `exploratory`, `observational`,
  `non-causal`, or a no-causality disclaimer to `association`, `relationship`,
  or `correlation`.
- Retain `exploratory` only when confirmatory versus post hoc status matters,
  and `observational` only when the study design or design contrast matters.
- Do not alter verified quantitative values, citations, formulas, or terminology.
- Do not hide a weak core argument under fluent prose. State material gaps plainly.
- If the central research question, evidence interpretation, or paper story needs
  substantive redesign, diagnose the problem and hand it to `academic-research`
  or `good-story`; do not solve it by silently changing the science during
  polishing.
- Prefer direct, information-dense claims over `we do not claim`, `this does not
  mean`, or `our purpose is not ... but ...`.
- Do not compress away a load-bearing noun head, referent, article, preposition,
  or scope phrase merely to shorten a sentence. When concise wording becomes
  underspecified, state the object or relation explicitly.
- Prefer conventional, field-appropriate academic collocations over novel,
  literal, or over-compressed phrasing when both preserve meaning. Natural
  disciplinary English takes priority over stylistic cleverness.
- Increase descriptive precision by naming the supported setting, population,
  measure, temporal or spatial resolution, or comparison when it materially
  helps interpretation. Greater specificity must not strengthen causality,
  generalisability, novelty, or mechanism beyond the evidence.
- Remove self-undermining comparison language such as `unfortunately`,
  `merely`, `only`, `still lags behind`, `failed to outperform`, or broad
  `performance degradation` when it turns a non-central comparison into a
  general weakness. State a material trade-off neutrally and specifically.
- Keep a limitation when removing it would overstate the evidence. Place it where
  it clarifies interpretation rather than leading a contribution-heavy sentence.
- Do not create a new weakness, comparator, or burden of proof during polishing.
  For whole-manuscript contribution, Results, or Discussion work, read
  `../shared/advantage-led-research-narrative.md`; reuse its continuity record,
  check Revision Invariants, preserve its integrity floor
  and do not hide a materially adverse result.
- Avoid em dashes as prose punctuation unless the user requests them.
- Use hyphens for compound modifiers and en dashes only for ranges or established
  forms. Do not create forms such as `mobility–based` or `indoor–outdoor`.

## Workflow

1. **Establish the target.** Identify the paper type, section, target journal or
   style, source language, requested depth, and whether the text is ready for
   `formal-polish` or `human-final-pass`.
2. **Diagnose before rewriting.** Check, in order:
   `paper type -> section job -> paragraph logic -> claim/evidence/boundary -> sentence style`.
3. **Load only the needed references.** Use the table below. Do not load every
   reference for a narrow sentence-level request.
4. **Rewrite at the highest necessary level.** Rebuild section or paragraph logic
   when needed; otherwise make the smallest wording change that solves the problem.
5. **Keep document-level state when needed.** For a whole manuscript, multiple
   sections, or repeated revision rounds, read `references/document-consistency.md`
   and maintain terminology and claim-strength ledgers.
6. **Preserve the evidence contract.** Keep claim strength aligned with the source,
   retain necessary uncertainty, and distinguish observation from interpretation.
7. **Run a final check.** Verify meaning, citations, numbers, terminology, section
   function, overclaim, defensive phrasing, self-undermining comparison,
   qualifier stacking, punctuation, and target-journal fit.

## Reference Router

| Reference | Read when |
|---|---|
| [references/writing-strategy.md](references/writing-strategy.md) | Section or paragraph reasoning needs repair |
| [references/published-article-patterns.md](references/published-article-patterns.md) | An abstract, Introduction, Results, Discussion, Conclusion, or title needs article-level patterns |
| [references/section-moves.md](references/section-moves.md) | A specific section needs rhetorical moves or ordering |
| [references/phrasebank-playbook.md](references/phrasebank-playbook.md) | Hedging, transitions, evidence, comparison, limitation, or implication wording is needed |
| [references/style-guardrails.md](references/style-guardrails.md) | Mechanics, register, articles, numbers, sentence checks, integrity, or AI boundaries need review |
| [references/defensive-phrasing.md](references/defensive-phrasing.md) | Defensive disclaimers must be retained, compressed, or rewritten |
| [references/document-consistency.md](references/document-consistency.md) | A whole manuscript, multiple sections, or repeated rounds need stable terminology and claim strength |
| [references/human-final-pass-policy.md](references/human-final-pass-policy.md) | `human-final-pass` is active |
| [references/human-final-pass-targeting.md](references/human-final-pass-targeting.md) | Section eligibility or protected content is uncertain |
| [references/human-final-pass-calibration.md](references/human-final-pass-calibration.md) | Naturalness intensity or explicit grammar-error mode needs calibration |

## Compact Section Check

- `Introduction`: importance -> known context -> unresolved gap -> question or aim
  -> approach and contribution, with a concise preview of the strongest supported
  finding. Keep detailed result narration in Results.
- `Results`: report what was observed, where or when, and with what quantitative
  support. Keep interpretation limited and intentional.
- `Discussion`: explain what the findings add, how they relate to prior work, which
  mechanism is plausible, and where the interpretation may fail.
- `Conclusion`: contribution -> key evidence -> memorable implication. Add no new
  data or defect; repeat a boundary only when needed to prevent misreading or
  required by the venue. Do not add a limitation or future-work quota.
- `Methods`: preserve reproducibility, assumptions, parameters, controls, software,
  and analysis details. Do not polish away necessary specificity.
- `Abstract`: context or problem -> gap or objective -> approach -> key results ->
  implication, unless the journal requires another structure.
- `Title`: accurate, searchable, evidence-supported, and no broader than the paper.

## Chinese-to-English Work

Extract propositions before translating. Reconstruct causal, contrastive,
qualifying, and inferential links instead of translating clause by clause. Keep
technical terminology stable and verify causality and hedging after reconstruction.

## Output

Unless the user requests another format, provide:

1. The polished text as plain prose.
2. `Revision notes:` with `3-5` short bullets covering substantive structural or
   stylistic changes.
3. A clear note when the rewrite changed section logic or exposed an unresolved
   evidence or argument gap.

For side-by-side requests, use `Original`, `Polished`, and `Why changed`.
For document-level work, report only unresolved terminology or claim-strength
conflicts unless the user asks to see the full ledgers.