domains / research / academic-paper-polish / paper-polish
The two skills share the name and the high-level goal of improving an academic manuscript, but they define different procedures. The HKUSTDial skill is a language-level polishing method focused on preserving the author's meaning, removing AI tone, and handling Chinese-to-English rewriting, with explicit rules against fabricating content and a conservative, single-pass workflow. The brycewang-stanford skill is a LaTeX-specific, 19-step proofreading and verification workflow that includes structural checks (titles, consistency, storyline, theoretical tension), a strict interactive approval loop for every issue, and file-scanning initialization. Neither copy is a reproduction or shortening of the other; crowning one would hide a real alternative approach to 'paper polish'.
| Repository | Path |
|---|---|
| HKUSTDial/Supervisor-Skills | skills/paper-polish/SKILL.md |
| brycewang-stanford/Auto-Empirical-Research-Skills | skills/67-econfin-workflow-toolkit/paper-polish/SKILL.md |
| Role | Repository | Path |
|---|---|---|
| peer | HKUSTDial/Supervisor-Skills | skills/paper-polish/SKILL.md |
| peer | brycewang-stanford/Auto-Empirical-Research-Skills | skills/67-econfin-workflow-toolkit/paper-polish/SKILL.md |
# Paper Polish ## Overview This skill polishes and verifies an existing academic manuscript. It supports two modes: a language-level polish that preserves the author's meaning, and a structured LaTeX proofreading and verification pass. Choose the mode based on the user's request and the manuscript format. ## When to Use - The user asks to polish, proofread, check, or improve an existing draft. - The user wants to remove AI tone, fix awkward wording, or translate a Chinese manuscript into publishable English. - The user has a LaTeX paper and wants a comprehensive check of titles, consistency, citations, formatting, and structural coherence. ## When NOT to Use - No prose exists yet and the user wants a section written from an idea. - The user wants the paper's logic skeleton planned from scratch. - The user wants a full reviewer-style audit beyond polishing and proofreading. ## Mode 1: Language-Level Polish ### Overriding Rule: Stay Faithful to the Author's Meaning - Never silently make an edit that changes scientific meaning. - Do not strengthen or weaken conclusions, turn correlations into causal claims, or drop qualifiers without explicit author confirmation. - Do not delete or heavily compress arguments; suggest cuts instead. - Pure language edits (grammar, word order, smoother phrasing with identical meaning) are safe to make freely. - Edits that might touch meaning must be flagged for the author to confirm. ### Never Fabricate Content - Add no data, numbers, equations, results, citations, or method names not in the original. - If a passage lacks evidence, leave a parenthetical note to the author instead of inventing content. ### Conservative by Default - Prefer the small edit over the big one, and no edit over the small one. - Do not "improve" hedged language into stronger claims. - If a structural rewrite is needed, stop and let the author choose between light polish and deep rewrite. ### Polishing Pass 1. Understand the passage's purpose and role in the paper. 2. Identify whether the problem is surface grammar or structural logic. 3. Edit within the faithfulness rule, applying section-specific conventions. 4. Deliver clean polished text and briefly state the main edits, flagging any that might have touched meaning. ### Section Conventions - Introduction: positions the work. - Methods: enables reproduction. - Results: states observations in past tense. - Discussion: interprets and hedges. - Abstract: miniature paper. - Title: searchable and restrained. ### AI-Tone Removal and Chinese-to-English Rewriting - Strip AI-generated phrasing and make the prose idiomatic. - For Chinese drafts, rewrite into submission-quality English while preserving the author's meaning and hedging. ## Mode 2: LaTeX Proofreading and Verification ### Initialization 1. Receive the paper directory path. 2. Scan the directory to identify the main LaTeX file, bibliography file, results directory, and any PDF with original tables/figures. 3. Read all files into context. 4. Present an inventory and confirm with the user before proceeding. ### Interaction Rule For every issue found in every check: 1. Present the issue clearly (original vs. proposed change, or finding). 2. Ask the user: "Accept this change? (yes/no/edit)" 3. Wait for the user's response before proceeding to the next issue. 4. If "yes", apply the change. 5. If "no", skip and move to the next issue. 6. If "edit" or alternative text is provided, apply the user's version. 7. After all issues in a check are resolved, announce completion and move to the next check. ### The 19 Checks Run these checks in order: 1. Section and Subsection Titles: Optimize titles for academic rigor; no colons. 2. Table and Figure Titles: Optimize captions for academic rigor; no colons. 3. Internal Consistency of Arguments: Check for contradictory or inconsistent claims across sections. 4. Introduction Logic and Flow: Ensure tight flow from motivation to findings to contributions. 5. Storyline Coherence: Evaluate overall storyline; suggest moving non-essential tables to appendix. 6. Theoretical Tension in Introduction Motivation: Ensure motivation establishes competing forces or perspectives that make the research question interesting ex ante. 7. Concise Results Reporting: Check that results are reported concisely and without redundancy. 8. Cross-Section Repetition: Identify and reduce unnecessary repetition across sections. 9. Em-Dash Usage: Standardize em-dash usage per style guide. 10. Auxiliary-Text-to-Footnote Conversion: Convert auxiliary text to footnotes where appropriate. 11. Citation Formatting: Verify citation format and consistency. 12. Reference List Completeness: Ensure all in-text citations appear in the bibliography and vice versa. 13. Equation Numbering and Referencing: Check equation numbering and cross-references. 14. Table and Figure Numbering: Verify sequential numbering and correct references. 15. Spelling and Grammar: Run a final pass for spelling and grammar errors. 16. Formatting Consistency: Check for consistent use of bold, italics, and other formatting. 17. Margin and Layout Issues: Identify any layout problems that may affect compilation or readability. 18. Compilation Warnings: Review LaTeX compilation log for warnings and errors. 19. Final Read-Through: Perform a final holistic read for flow and clarity. ## Conflicts - HKUSTDial/Supervisor-Skills::skills/paper-polish/SKILL.md does not mandate a per-issue interactive approval loop, while brycewang-stanford/Auto-Empirical-Research-Skills::skills/67-econfin-workflow-toolkit/paper-polish/SKILL.md requires it for every issue in every check. - HKUSTDial/Supervisor-Skills::skills/paper-polish/SKILL.md includes Chinese-to-English rewriting and AI-tone removal, which are not part of the brycewang-stanford/Auto-Empirical-Research-Skills::skills/67-econfin-workflow-toolkit/paper-polish/SKILL.md procedure. - brycewang-stanford/Auto-Empirical-Research-Skills::skills/67-econfin-workflow-toolkit/paper-polish/SKILL.md includes LaTeX-specific checks (e.g., em-dash usage, auxiliary-text-to-footnote conversion, compilation warnings) that are absent from the HKUSTDial/Supervisor-Skills::skills/paper-polish/SKILL.md procedure.