Writing Polish

---
name: writing-polish
description: >-
  Polishing of the USER'S OWN academic prose across five dimensions — accuracy,
  clarity/readability, formality, register & conventions, and stance — for language
  assessment / applied linguistics writing. Built for multilingual and early-career
  academics. Use when the user asks to "polish my writing", "improve the English",
  "check the style / register / formality", "is this too informal", or wants a draft
  to read like the field's published prose. Accuracy errors are corrected directly
  (with a change log); everything stylistic is numbered proposals the author accepts
  one at a time, each with its reason — preserving the author's voice throughout.
---

# writing-polish

Polishes the user's own academic prose at two speeds: unambiguous **accuracy errors are corrected directly** and reported in a change log, while everything that touches style or voice arrives as **numbered proposals** the author accepts one at a time, each with a one-line reason. The reason is the teaching: the goal is that the writer needs this skill less each time. It improves writing the author produced; it does not produce writing.

**Boundary and disclosure — read first.**

- **The user's own draft only.** This skill exists to help multilingual and inexperienced academics write to the field's linguistic and conventional standards. It is designed to refine the language of the user's own work while preserving their authorial voice — the ideas and claims stay theirs; the polish only makes them clearer.
- **Language polish is the AI use most publishers explicitly permit** — several allow "language improvement" assistance outright — but permission is not exemption: remind the user, every run, to check the target venue's AI policy and include whatever declaration it asks for. What to declare and how is the author's call.
- **Outputs quote the draft**: treat them as part of the manuscript's confidential record — keep them beside it, never in a public repository.

## Step 0 — intake

Establish: the **venue** (load its section from `../pre-submission-review/references/journal-reviewer-guides.md` when listed — register expectations differ, and spelling variety follows the venue); the **section/genre** being polished (an abstract, a methods section, and a discussion carry different conventions — tense, voice, hedging norms); and **scope** (whole draft or excerpt). Ask only what the text doesn't answer.

## The preserve-list (charter — checked mechanically, never waived)

Never change: **numbers and statistics** · **citations, citekeys, and the reference list** · **direct quotes** · **technical terms and construct names** · **meaning** · **epistemic commitment** (how strongly a claim is made). Where a prose problem is entangled with claim strength, do not silently fix it — raise a flagged **proportionality proposal** with the rationale (see dimension E), and point the user at the `pre-submission-review` skill if such findings cluster: that is review territory, not polish.

## The five dimensions

Work through them in order; report proposals grouped by dimension.

**A · Accuracy** — the language itself: grammar, word form, tense, articles, prepositions, collocations, countability, mechanics. **This dimension is corrected directly**: unambiguous mechanical errors don't deserve one-at-a-time approvals — fix them, and deliver a concise change log (original → corrected). Two guards: the preserve-list still governs every correction, and anything where the fix could shift meaning — or where two corrections are equally plausible — is instead shown as a **tracked-changes-style suggestion** (marked deletion → insertion, in place) for the author to accept or decline. The field's own standard applies to us: the register of professional editing, never "a native speaker would say…".

**B · Clarity and readability** — one idea per sentence; keep subject and verb close; prefer verbs to nominalizations ("we analysed" over "the analysis was conducted"); resolve ambiguous references ("this" pointing at nothing); cut inflated openers, boilerplate ("it is worth noting that"), and doubled expressions. **Watch especially for unnecessarily long, complicated sentences — and above all for *clusters* of them**: one dense sentence is a style choice, three in a row is a wall; propose breaking the cluster and varying sentence length so the paragraph breathes. Every cut is meaning-neutral by the preserve-list.

**C · Formality** — colloquialisms ("what's more" → "moreover"; "a lot of" → "considerable"), contractions, conversational connectors ("so", "besides" as sentence openers), emotive intensifiers ("hugely important"), vague quantifiers ("some researchers say") — while avoiding the opposite failure: over-formalizing for formality's sake. Formal words like "utilize" or "commence" are perfectly legitimate; the failure is choosing them *because* they sound formal when "use" or "begin" is equally precise, and doing so habitually until the prose stiffens. Propose the plain form only where the longer one adds nothing — and never where the longer form is the field's own term of art ("rating-scale category utilization" stays).

**D · Register and conventions** — the field's register is **objective and cordial**: about the evidence, never persons; critical without being combative. Genre conventions: topic sentences that promise what the paragraph delivers; cohesion signals between moves; aims stated where readers of this genre expect them; tense conventions by section (established findings vs. the present study); first-person use per the venue's own norms.

**E · Stance** — the interpersonal layer of academic writing: hedges ("may", "appears to"), boosters ("clearly", "demonstrates"), attitude markers, and self-mention. Stance is **conventions territory with teeth**, so it gets its own pass and its own rule. What the skill *proposes freely*: stance-appropriateness edits that leave commitment unchanged — a hedge in the genre's expected form, a booster moved to where the evidence supports it, self-mention aligned with venue norms. Stance is **move-sensitive** (conclusions expect calibrated caution; results statements may be confident) and **discipline-sensitive** (repertoires differ by field). What the skill *never does silently*: insert or delete a hedge or booster in a way that strengthens or weakens the claim itself — that changes epistemic commitment, and it is always a flagged **proportionality proposal** with the rationale, the author deciding. For writers building their stance repertoire, point them at the University of Manchester's **[Academic Phrasebank](https://www.phrasebank.manchester.ac.uk/)** (free) — especially its *Being cautious* and *Being critical* sections.

## Output — proposals, not a rewrite

Two layers. **Accuracy (A): the corrected text plus a concise change log** (original → corrected), with **meaning-risk items shown as tracked-changes-style suggestions in place** — marked deletion → insertion — so the author decides each one where it stands (for Word drafts: deliver the edited copy and note that Word's *Compare* turns it into true tracked changes). **Style (B–E): numbered proposals** grouped by dimension — **original → proposed → one-line reason** — the author accepts or rejects each. **Minimal-intervention rule** throughout: the smallest edit that fixes the problem — the author's voice is not replaced with a house voice, and text with no problem is left alone (a competent paragraph gets no edits, and that is not a failure). A clean full version is assembled only on request: accuracy corrections plus accepted proposals, nothing else. Close with a **recurring-patterns note** (at most three, by dimension), naming each pattern and **signposting the matching free resource** — a stance pattern points at the Academic Phrasebank's *Being cautious*/*Being critical* section, an accuracy pattern at Purdue OWL's multilingual-writers pages, a style/formality pattern at UEFAP — so the writer takes away their own personal checklist with somewhere to practise it. The teaching lives here, not in the change log.

## Self-check (always, before presenting)

Verify mechanically that every preserve-list item is intact between original and proposal — numbers, citations, and quotes byte-identical; claim strength and hedging unchanged except where a proportionality proposal is flagged as such; no proposal changes meaning. A polish that quietly moved a number or firmed up a claim is not a polish; it is an error this check exists to catch.

## References

- `../pre-submission-review/references/journal-reviewer-guides.md` — venue register expectations and the publishers' own tone canons (the objective/cordial standard is theirs as much as ours).
- `../../docs/04-citation-integrity.md` — the charter; this skill inherits its declaration rules and touches no citations.
- [Academic Phrasebank](https://www.phrasebank.manchester.ac.uk/) (University of Manchester; free to use online, content © the university — link and recommend, never copy its lists into files) — the writer-facing companion for stance and phraseology, especially *Being cautious* and *Being critical*.
- [Purdue OWL](https://owl.purdue.edu/owl/index.html) (free, open-access) — sentence-level clarity, mechanics, and a dedicated multilingual/ESL writers collection; the companion for dimension A patterns.
- [UEFAP — Using English for Academic Purposes](https://www.uefap.org/) (Andy Gillett; free) — EAP-specific guidance across the skill areas; strong on academic style and functions, the companion for dimensions C–D.

## Provenance

Built on what the field's publishers explicitly permit (language-improvement assistance, with disclosure), on the register standard this toolkit's review skills already carry (objective · clear · cordial), and on the field's own analytic tradition for these dimensions — language accuracy and academic conventions/stance are the very layers the field's writing scales separate, hedging/boosting/self-mention follow its standard metadiscourse framework, and the dimensions align with the features of academic writing long taught in EAP (formality, precision, objectivity, explicitness, caution). Prompted by community clarity-editing tools (MIT); rebuilt fresh to this toolkit's charter — the propose-only diffs, the mechanical preserve-list, and the stance proportionality flags are the non-negotiable differences, and the framing is deliberately what it is: better academic writing for the scholars who write it, not camouflage for text nobody wrote.