AI Skill Report Card
Editing Scientific Texts
YAML--- name: editing-scientific-texts description: Reviews scientific manuscripts and texts, providing structured feedback on scientific accuracy, writing style, and text composition, plus direct grammar/spelling corrections. Use when asked to edit, proofread, review, or lector a scientific document, paper, thesis, or report. --- # Editing Scientific Texts
Quick Start14 / 15
Given a scientific document, return it interspersed with inline comments and, where needed, rewritten section suggestions. Corrections to grammar/spelling are shown directly in the text using strikethrough + replacement; substantive comments go in bracketed annotations.
Annotation format:
The results demonstrates ~~demonstrates~~ **demonstrate** a clear correlation.
[COMPOSITION: This paragraph introduces a new claim about causality without
transitioning from the previous paragraph's discussion of correlation. Consider
adding a linking sentence, e.g., "While the above shows correlation, the
following experiment tests causality directly."]
[SCIENTIFIC ACCURACY: The claim that "X always leads to Y" is too strong given
n=12 and no control group. Suggest: "X was associated with Y in this small
sample, though causal direction cannot be established."]
[STYLE: This sentence runs 58 words with four subordinate clauses. Consider
splitting into two sentences for readability.]
For a section requiring substantial rework, provide:
[REWRITE SUGGESTION — Section 2.3]
Original: [quote or summarize]
Suggested rewrite:
"[new version]"
Rationale: [why the rewrite improves clarity/accuracy/flow]
Recommendation▾
Add an example showing a full document pass with multiple comment categories interacting, to demonstrate final summary compilation.
Workflow14 / 15
Progress:
- Read the full text once without commenting, to grasp argument and structure
- Pass 1 — Scientific accuracy: check claims, terminology, logical consistency, evidence-conclusion fit
- Pass 2 — Composition: check overall thread, section ordering, paragraph-to-paragraph flow, whether intro/conclusion match body
- Pass 3 — Writing style: sentence length/variety, sentence-to-sentence flow, redundancy, passive voice overuse
- Pass 4 — Grammar & spelling: mark direct corrections inline
- Compile all annotations into the returned document, preserving original text with markup
- Write a brief summary at the end (3–6 bullets) of the most important issues across all categories
Recommendation▾
Include guidance on handling non-English or translated scientific texts, or citation/reference formatting issues, which are common in manuscript editing.
The Four Comment Categories
- Understandability — Can a reader in the field follow this without re-reading? Flag jargon left undefined, ambiguous referents ("this shows..." — what does "this" refer to?), missing context.
- Writing style — Sentence length (flag >35 words), sentence variety, transitions between sentences, unnecessary hedging or nominalizations, awkward phrasing.
- Text composition — Is there a clear thread/argument line? Does each section build logically on the last? Do intro and conclusion align with the body? Are paragraphs internally unified (one idea each)?
- Grammar & spelling — Direct, unambiguous corrections marked with strikethrough + replacement. No discussion needed unless the correction changes meaning.
Examples17 / 20
Example 1: Input: "The data shows that the reaction occur faster at higher temperature, this support our hypothesis."
Output:
The data shows ~~shows~~ **show** that the reaction occur ~~occur~~ **occurs**
faster at higher temperature ~~temperature, this~~ **temperatures. This** supports
~~support~~ **supports** our hypothesis.
[GRAMMAR: "data" is plural (datum/data), and the comma splice was corrected
by splitting into two sentences.]
Example 2: Input: A methods section that jumps from describing sample collection directly into statistical results without stating which test was used or why.
Output:
[SCIENTIFIC ACCURACY: The statistical test applied is not named before results
are reported. State the test (e.g., "A two-tailed t-test was used to compare...")
before presenting p-values, so readers can judge appropriateness.]
[COMPOSITION: There is a gap between "Sample Collection" and "Results" —
readers expect a "Statistical Analysis" subsection bridging the two. Consider
adding one.]
Example 3: Input: A 62-word sentence with five clauses buried in the discussion section.
Output:
[STYLE: This sentence (62 words, 5 clauses) asks a lot of the reader. Suggest
splitting at the semicolon and again after the second "which," e.g.:
"X was observed in all trials. This suggests Y, which is consistent with Z.
However, the effect size was small, which may limit generalizability."]
Recommendation▾
Clarify how to handle conflicting priorities (e.g., accuracy vs. style) when time/space for feedback is limited.
Best Practices
- Preserve the author's voice; suggest, don't rewrite wholesale unless asked or unless the passage is genuinely unclear.
- Always give a reason for style/composition/accuracy comments — never just "awkward" or "unclear," explain what's awkward and why.
- For scientific accuracy comments, be specific about what evidence would be needed to support a stronger claim.
- Keep grammar corrections silent/inline (no lengthy explanation) unless the fix changes meaning.
- End with a short prioritized summary — editors should know what to fix first.
- Match tone to the register of the field (a physics paper and a medical case report tolerate different levels of formality).
Common Pitfalls
- Don't conflate a style preference with a grammar error — mark style suggestions as comments, not strikethroughs.
- Don't rewrite entire paragraphs by default; reserve full rewrites for passages that are genuinely broken, and always explain the rationale.
- Don't give vague feedback like "improve flow" — point to the specific transition or sentence causing the break.
- Don't ignore the paper's overall argument thread while focused on line edits — always check that local fixes still serve the composition-level goal.
- Don't overcorrect valid field-specific jargon or conventions as if they were errors.