cgpt-academic-proofreader

name	cgpt-academic-proofreader
description	Report-only academic manuscript proofreading for grammar, journal style, argument flow, terminology, citation mechanics, statistical reporting, and Quarto cross-reference readiness; use before sharing, submission, or after manuscript revisions.

CGPT Academic Proofreader

Use this skill when the user asks for proofreading, copyediting review, pre-submission language checks, journal-readiness review, or a manuscript quality report. The default mode is report-only: identify issues and propose fixes, but do not edit manuscript files.

Default Behavior

• Report only unless the user explicitly asks to apply approved fixes.
• Do not edit .qmd, .md, .docx, .bib, figures, tables, or project logs during the review pass.
• Write a report only when the user asks for a file-based review or the project already uses quality_reports/.
• Prioritize peer-review risks over cosmetic wording.
• Coordinate with cgpt-stats-audit for numerical verification and cgpt-citation-bib-audit for deep bibliography checks. Do not duplicate full stats or external citation audits unless requested.

Setup

Before reviewing, inspect only the relevant project guidance:

• target manuscript or section
• CLAUDE.md, MEMORY.md, SESSION-LOG.md, or project notes if present
• .bib files when citation mechanics or key resolution are in scope
• _quarto.yml or master QMD include chains when cross-references or rendered structure matter

If exact line numbers are available, include them. If reviewing DOCX or rendered text without stable lines, use section, paragraph, heading, or page context.

Review Checklist

Check every requested manuscript for:

• grammar, agreement, articles, tense, sentence boundaries, and informal contractions
• journal tone, excessive hedging, weak openings, and unclear topic sentences
• causal language in observational work, including verbs such as "causes," "drives," "impacts," and "mediates" when not warranted
• author-first literature summaries that should be claim-first
• paragraph structure: claim, evidence, synthesis, implication or transition
• terminology, symbols, acronym definitions, variable names, and sample labels
• statistical reporting completeness at the prose level, including sample sizes, estimators, uncertainty, p-values, effect sizes, and confidence intervals where applicable
• citation mechanics, including unresolved @key patterns, inconsistent narrative versus parenthetical syntax, and likely missing support for factual claims
• Quarto references such as @fig-, @tbl-, @eq-, and @sec-
• captions, table and figure callouts, heading hierarchy, abstract presence, and obvious completeness gaps

Treat unresolved citations, unsupported factual claims, causal overreach, missing sample definitions, and broken cross-references as high-priority risks.

Severity

• Critical: likely to block rendering, submission, or citation integrity.
• High: likely reviewer or editor concern, including causal overclaiming, unsupported claims, missing sample definitions, major argument gaps, or unresolved references.
• Medium: clarity, consistency, style, or reporting issues that weaken the manuscript.
• Low: local grammar, phrasing, punctuation, or polish.

Report Format

Use this structure for a full review:

# Academic Proofreading Report
**Date:**
**Document:**
**Mode:** Report only

## Summary
- Critical:
- High:
- Medium:
- Low:
- Overall readiness:

## Must Fix Before Sharing or Submission
| Location | Severity | Issue | Proposed fix |
|---|---|---|---|

## Detailed Issues
| Location | Category | Severity | Current text | Proposed text | Rationale |
|---|---|---|---|---|---|

## Cross-Checks
| Check | Result | Notes |
|---|---|---|

## Overall Assessment
- Strengths:
- Priority fixes:
- Recommended next audit:

For a narrow proofread, keep the same severity logic but report only the requested sections or issue categories.

Applying Fixes

Only apply changes after explicit approval or when the user directly asks for edits. Then:

1. Apply only approved fixes.
2. Keep edits scoped to manuscript text, not data or analysis logic.
3. Re-read the affected passages.
4. If Quarto, DOCX, citations, or cross-references may be affected, recommend or run the appropriate render/check skill when in scope.

If approval is ambiguous, stop at the report.