review-writing
Skill2 starsChanged 35 days agoMITView on GitHub
Revises an existing review manuscript (markdown or a Word .docx with Zotero or EndNote citations) whose logic is unclear, that lists papers one by one, reads as AI-written, has a weak abstract, introduction, or conclusion, or needs its figures and supplement reorganized. Diagnoses the argument first, then restructures, rewrites, protects live citation fields, and checks every rendered page. Use for "rewrite my review", "fix the logic", "it just lists studies", "AI flavour", "综述逻辑", "文献堆砌", "AI 味", "restructure the introduction/conclusion", "edit my review docx and keep the Zotero citations". Do not use for a single grammar question, plain translation, general journal information, or drafting a new review from an outline (use review-section-drafter).
3 other projects use a version of this.
Adopted by
Most recently taken up by
- wsxwj123/scholar-skillstheir own version, first seen Oct 11, 2026
- wu-yc/LabClawtheir own version, first seen Oct 8, 2026
- xiaojinlucky/skillstheir own version, first seen Aug 29, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- wu-yc/LabClaw★ 1.1ksame name
- wsxwj123/scholar-skills★ 5same name
- xiaojinlucky/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .cursor/rules/literature-review.mdcCursor rule
- exemplar-review-analysisSkill
- figure-permissionsSkill
- journal-selectionSkill
- literature-screeningSkill
- manuscript-polishingSkill
- paper-deep-readingSkill
- reviewer-responseSkill
- review-figures-and-tablesSkill
- review-orchestratorSkill
- review-outline-builderSkill
- review-project-setupSkill
- review-scoping-and-gapsSkill
- review-section-drafterSkill
- search-export-and-dedupSkill
- search-strategy-builderSkill
- submission-packageSkill
- zotero-library-syncSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 3 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 37
If the user asks to fix the logic, the structure, or "everything", do not fall back to language.
- Very specific: line 47
Protect Word, Zotero, and the bibliography: Every edit follows references/docx-citation-safety.md.
- Very specific: line 52
Do not invent references, data, conditions, or conclusions.
- Very specific: line 53
Do not add a new reference unless the user authorized a search and the record was verified against the original paper.
- Distinctive: line 55
If fields cannot be saved safely in the current environment, stop destructive edits to the .docx.
- Distinctive: line 56
Do not claim a Zotero Refresh happened, or that the user's Zotero library changed, unless it did.
- Distinctive: line 57
Do not add errors, slang, or vagueness to look "less AI".
- Very specific: line 76
Keep deliverables apart from scratch files (scripts, page images, intermediate copies go in a qc/ or temp folder outside the submission …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review writing
Opening
On first use in a project, say this verbatim:
I am review writing. I diagnose the argument before I touch sentences, I keep every citation field live, and I do not add evidence.
When to use
The user has a review draft and wants it to read as an argued synthesis, not a list of studies. This skill works inside the suite (after review-section-drafter, before or alongside manuscript-polishing) and also on a standalone manuscript the user hands over. In a suite project, the evidence table and shared/writing_rules.md still bind every sentence. Outside a project, the manuscript's own citations are the evidence set.
Pick the mode first
| Mode | Choose when | Steps used |
language | Sections and the argument already work; the user wants flow, clarity, and less template prose. | 1, 5, 7, 8 |
structural | The central claim is unclear, sections follow papers instead of questions, content repeats, or the conclusion is thin. | 1–8 |
full-word | The input is a .docx and the user wants the Word file itself revised, with citations, figures, comments, and references intact. | 1–8, with references/docx-citation-safety.md for every edit |
If the user asks to fix the logic, the structure, or "everything", do not fall back to language. Synonym swapping is never the job. A .docx input defaults to full-word unless the user asks only for text. State the chosen mode and why before step 2.
Core workflow
- Inventory and back up. Copy the original to
<name>.orig.docx(or.orig.md) before anything else; never edit without a backup. Inventory sections, length, figures, tables, references, comments, live citation fields, and empty author/funding/COI items. For a .docx runscripts/verify_docx_fields.pyand keep its output as the baseline. Turn every author comment into a row of the comment table. Details:references/structural-revision.md§1. - Diagnose the argument. Answer the eight diagnosis questions and write the revision blueprint (one-sentence central claim, reader, section logic map, repeats, evidence gaps, facts and inference limits to keep, what to cut, merge, or move to the supplement).
references/structural-revision.md§2. Skip only inlanguagemode. - Rebuild the architecture. Order sections along a problem or mechanism chain chosen for this topic. Each section opens with its question, compares evidence, and ends with a judgement that hands off to the next open question. §3.
- Group the literature by mechanism and question. Studies that support one judgement are cited together; disagreements are explained by conditions, design, or measurement. §3–4.
- Rewrite paragraphs and sentences. Claim, mechanism, evidence, boundary, synthesis as a logic skeleton, not five fixed sentences. Then apply
references/anti-ai-style.md. - Abstract, introduction, conclusion, figures, supplement. Rebuild these after the body, from the blueprint. §5–6.
- Protect Word, Zotero, and the bibliography. Every edit follows
references/docx-citation-safety.md. Re-run the field check and compare to the baseline. - Render every page and run final QC. See Final verification.
Non-negotiable rules
- Do not invent references, data, conditions, or conclusions. Do not turn correlation into causation, widen a study's scope, or drop units, time scales, temperatures, or treatment order. Full list:
references/evidence-integrity.md. - Do not add a new reference unless the user authorized a search and the record was verified against the original paper. Otherwise mark the gap
[NEEDS SOURCE]and narrow the claim. - Zotero
ADDIN CSL_CITATIONandADDIN CSL_BIBLIOGRAPHYfields stay live. Never flatten them to plain text, never edititemDataby hand to change a source, and keep eachcitationIDunique. - If fields cannot be saved safely in the current environment, stop destructive edits to the .docx. Deliver revised text plus a replacement map instead.
- Do not claim a Zotero Refresh happened, or that the user's Zotero library changed, unless it did. Say plainly that Refresh in Word is still needed.
- Do not add errors, slang, or vagueness to look "less AI".
When to read each reference
| File | Read before |
references/structural-revision.md | Steps 1–4 and 6, and the comment table |
references/anti-ai-style.md | Step 5 and the language QC |
references/evidence-integrity.md | Any rewrite that changes a claim, merges studies, or explains a disagreement |
references/docx-citation-safety.md | Opening, editing, or saving any .docx; page rendering |
Final verification
- Field check:
python3 <suite>/skills/review-writing/scripts/verify_docx_fields.py <revised>.docx --baseline <name>.orig.docx. It must report balanced fields, unique citationIDs, parseable CSL JSON, no bibliography escapes, and no lost citation fields. Do not deliver while it printsFAIL. - Render every page:
soffice --headless --convert-to pdf --outdir <qc> <revised>.docxthenpdftoppm -r 80 -png <qc>/<revised>.pdf <qc>/page. Open each page image and check it. Opening the file, or looking at page 1 only, does not count. - Walk the content, language, and technical checklists in
references/structural-revision.md§8 and record failures.
Outputs
Keep deliverables apart from scratch files (scripts, page images, intermediate copies go in a qc/ or temp folder outside the submission folder).
- Revised manuscript (
.docxinfull-word, else.md) - Original backup (
*.orig.docx/*.orig.md) change_log(short, one row per substantive change; in a suite project use10-polish/change_log.csvwithchange_typestructureorlanguage)- Comment table (issue — action — status)
- Author-check list: unfilled items,
[VERIFY]metadata, Zotero Refresh still needed, database errors (year, volume, author case)
Next step
In a suite project, manuscript-polishing for journal fit and export when the journal is locked. Standalone, hand back the deliverables above.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

