paper-self-revise
Revise an academic paper based on internal review comments. Reads review report or annotated manuscript, then applies revisions one by one with user approval. Trigger when user says "self revise" / "paper-selfrevise" / "内部修改" / "根据审稿意见修改".
Install / Use
npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill paper-self-reviseInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
MarketingSupported Platforms
Our assessment of paper-self-revise
paper-self-revise scores 95/100 on our quality scale, 80th of 642 Marketing skills we index (top 13%).
Its SKILL.md is 13 KB long, well organised into 18 sections with 7 code examples: a thorough specification that gives an agent plenty to work with.
With 4,360 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated about 2 months ago, so paper-self-revise is actively maintained.
- Our last check on 2026-11-15 found the source still online.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 88/100, with 1 caution from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-09-29. Automated pattern scan on 2026-09-29. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
paper-self-revise compared with similar skills
All 4 of these similar skills score higher than paper-self-revise; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| paper-self-revise (this skill)by brycewang-stanford | 95 | 4.4k | 54d ago | SKILL.md |
| x402by browser-use | 100 | 121.0k | 3d ago | SKILL.md |
| algorithmic-artby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
| pptxby anthropics | 100 | 177.9k | 55d ago | SKILL.md |
| designby nextlevelbuilder | 100 | 149.0k | 6d ago | SKILL.md |
Frequently asked questions
- How do I install paper-self-revise?
- Run
npx skills add brycewang-stanford/Auto-Empirical-Research-Skills --skill paper-self-revise. The install tabs above show the steps for each supported agent. - Which AI agents does paper-self-revise work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is paper-self-revise safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It declares no license and scores 88/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is paper-self-revise still maintained?
- The repository was last updated about 2 months ago, so paper-self-revise is actively maintained.
Skill content
View source on GitHubname: paper-self-revise description: Revise an academic paper based on internal review comments. Reads review report or annotated manuscript, then applies revisions one by one with user approval. Trigger when user says "self revise" / "paper-selfrevise" / "内部修改" / "根据审稿意见修改". allowed-tools: Read, Edit, Write, Glob, Grep, AskUserQuestion, Bash argument-hint: "[path-to-project-folder]"
Paper Self-Revise
Overview
This skill revises an academic paper (main.tex) based on internal review comments. The user provides a project folder containing the manuscript and review materials. The skill extracts review comments, then processes them one by one: displaying the comment, showing the original text, proposing a revision, and asking for user approval before applying the change.
Input: A project folder containing:
main.tex— the main manuscript (the file to be modified)- Subfolders with figures/tables (referenced by the manuscript)
- Review materials (one or more of the following):
- A review report file (e.g.,
review.txt,review.md,review.pdf,report.txt,report.md,report.pdf) - An annotated manuscript (e.g.,
annotated.tex,annotated.pdf) where annotations/comments serve as review comments
- A review report file (e.g.,
Output: The revised main.tex with all approved changes applied in place.
Workflow
Phase 0: Review Report Check
- Receive folder path from user (via
$ARGUMENTSor ask). - Ask the user: "您是否已有现成的审稿报告(review report)?(yes/no)"
- If yes: Proceed to Phase 1.
- If no: Draft a review report before proceeding:
a. Read
main.texand all supporting materials (tables, figures) in the project folder to fully understand the paper. b. Draft a review report in English (~3000 words), structured as:- A brief overall assessment (1–2 paragraphs)
- Numbered comments: Comment 1, Comment 2, Comment 3, etc.
- Each comment should identify a specific issue, explain why it matters, and suggest how to address it.
- IMPORTANT: Exclude any comments related to empirical tests that would require changing software code (e.g., re-running regressions, adding new specifications, changing estimation methods, running robustness checks in Stata/R/Python). Focus only on writing, argumentation, framing, literature, structure, clarity, interpretation, and presentation.
c. Save the review report as
review.pdfin the project folder using a Bash command to generate the PDF (e.g., via pandoc or a Python script with reportlab/fpdf). If pandoc is available, first write the content toreview.mdthen convert:pandoc review.md -o review.pdf. If pandoc is not available, use Python to generate the PDF. d. Count the words of the generated review report (count words in the source text — thereview.mdcontent or the raw English text used to buildreview.pdf, excluding markdown/PDF formatting markup). Use a Bash command such aswc -w review.mdor a Python word-count snippet on the text. e. Inform the user, displaying the actual word count: "审稿报告已生成并保存为 review.pdf(共 X 词),现在开始逐条修改。" — substituteXwith the measured count. f. Proceed to Phase 1 using the newly createdreview.pdfas the review material.
Phase 1: Initialization
- Scan the folder using Glob to identify all files:
- Locate
main.tex(the target file to revise). - Locate review materials by searching for files matching common review file patterns:
review*,report*,referee*,comments*,annotated*,feedback*,revision*. - If multiple review files are found, list them and ask the user which one(s) to use.
- If no review file is found, ask the user to specify the review file path.
- Locate
- Read
main.texin full to understand the current manuscript (skip if already read in Phase 0). - Read the review file(s):
- For
.texfiles: look for comments marked with%or annotation commands (e.g.,\annotate{},\comment{},\todo{}, or any custom annotation commands). - For
.txt/.mdfiles: read the full content as review comments. - For
.pdffiles: read the PDF content to extract review text. - For
.docxfiles: use Bash to extract text content.
- For
- Detect tracked changes (revision marks) in the annotated manuscript:
- Look for revision markup such as strikethrough/deletion marks, inserted text, colored text indicating additions/deletions, or LaTeX track-changes commands (e.g.,
\added{},\deleted{},\replaced{}{},\st{},\sout{},\hl{}, or similar commands from packages likechanges,trackchanges,latexdiff). - Tracked changes in the body text are revision items just like annotation comments — each one must be presented to the user for approval in Phase 3.
- Look for revision markup such as strikethrough/deletion marks, inserted text, colored text indicating additions/deletions, or LaTeX track-changes commands (e.g.,
Phase 2: Extract and Structure Review Comments
- Parse the review materials to extract individual review comments/suggestions. This includes two categories:
- Annotation comments: margin comments,
%-comments,\todo{},\comment{}, etc. - Tracked changes: inline revision marks (insertions, deletions, replacements) in the body text. Each distinct tracked change is treated as a separate revision item, equivalent to an annotation comment.
- Annotation comments: margin comments,
- Filter out formatting-only comments: Automatically discard any comments or tracked changes that are purely about formatting (e.g., font size, spacing, margins, line breaks, paragraph indentation, column layout, page layout, figure/table placement commands). Do NOT present these to the user. Only retain comments about substantive content (wording, argumentation, data, analysis, structure, logic, references, etc.).
- Structure each comment as a discrete revision item with:
- Comment number (sequential)
- Type — either "批注" (annotation comment) or "修订痕迹" (tracked change)
- Review comment text (the reviewer's original words, or for tracked changes: a description of what was deleted/inserted/replaced)
- Location in manuscript (section, paragraph, or specific text the comment refers to)
- Present the full list of comments to the user:
I have extracted the following review comments:
No. | Type | Location | Comment Summary
----|----------|---------------------------|------------------------------------------
1 | 批注 | Section 3, para 2 | Clarify the identification strategy
2 | 修订痕迹 | Section 4.1, Table 2 | Replace "effect" with "impact" in sentence about...
3 | 批注 | Section 5, para 1 | Strengthen the mechanism discussion
...
Total: N comments. Shall I proceed with revisions one by one?
- Wait for user confirmation before proceeding.
Phase 3: Iterative Revision (Core Loop)
For each review comment, follow this exact interaction pattern:
Step 1: Display the review comment
For annotation comments (批注):
───────────────────────────────────────────
批注 [X/N]:
[Full text of the reviewer's comment]
───────────────────────────────────────────
For tracked changes (修订痕迹):
───────────────────────────────────────────
修订痕迹 [X/N]:
[Description: e.g., "删除了 'XXX',替换为 'YYY'" or "插入了 'XXX'" or "删除了 'XXX'"]
───────────────────────────────────────────
Step 2: Show original text
Locate and display the relevant passage from main.tex:
Original text:
┌─────────────────────────────────────────┐
│ [The exact paragraph(s) from main.tex │
│ that this comment refers to] │
└─────────────────────────────────────────┘
If the comment refers to missing content (e.g., "add a discussion of X"), show the surrounding context where the new text should be inserted.
Step 3: Propose revision
For annotation comments: Generate a revised version of the text that addresses the reviewer's comment.
For tracked changes: The proposed revision is the result of accepting the tracked change (i.e., applying the deletion/insertion/replacement as marked by the reviewer).
Proposed revision:
┌─────────────────────────────────────────┐
│ [The revised paragraph(s) that address │
│ the reviewer's comment] │
└─────────────────────────────────────────┘
Rationale: [Brief explanation of what was changed and why]
Step 4: Ask for user decision
Use AskUserQuestion to ask:
Do you approve this revision? (yes / no / skip / edit)
- yes: Apply this revision to main.tex
- no: I will propose a new revision
- skip: Skip this comment and move to the next
- edit: You provide the revised text yourself
Step 5: Handle user response
- "yes" / "y" / "同意" / "好": Use the Edit tool to replace the original text with the proposed revision in
main.tex. Confirm the change was applied, then move to the next comment. - "no" / "n" / "不同意" / "不行": Ask the user what they would like changed about the proposed revision, then generate a new proposal. Return to Step 3.
- "skip" / "跳过": Skip this comment without making changes. Move to the next comment.
- "edit" / "我来改": Ask the user to provide their own revised text via AskUserQuestion, then apply that text using the Edit tool. Move to the next comment.
Phase 4: Completion
After processing all comments:
- Present a summary:
Revision complete!
Applied: X / N items (A annotation comments + B tracked changes)
Skipped: Y items
Auto-filtered (formatting-only): Z items
Total changes made to main.tex: X
Modified sections:
- Section 3: 2 revisions
- Section 4: 1 revision
- Section 5: 3 revisions
- Offer to re-read the revised
main.texto check for consistency or flow issues introduced by the revisions.
Writing Rules
Style and Formatting
- Maintain the same writing style as the original manuscript (do not change the overall tone or voice).
- Preserve all LaTeX formatting:
\section{},\subsection{},\ref{},\cite{},\citep{}, math environments, etc. - Preserve all labels (
\label{}), cross-references (\ref{}), and citations intact unless the review specifically asks to change them. - When adding new citations, use the same citation style as the rest of the paper (typically 30%
\cite{}/ 70%\citep{}). - When reporting regression results, use t-statistics in parentheses and "significant at X% level" — no p-values or confidence intervals.
- Avoid the em-dash (
---): Do not use the LaTeX em-dash---in revised or newly written prose. It is a common AI-writing tell and clutters academic text. Replace each occurrence with the punctuation its grammatical role calls for: a parenthetical aside becomes parentheses (or commas if short); a connector before an explanation becomes a colon or semicolon; an abrupt shift becomes a period and two sentences; a list continuation becomes a comma plus a connecting word. Leave the en-dash--used for page and number ranges untouched (e.g.,pages={259--271},2012--2026). When a revision touches a paragraph, also strip any pre-existing---from that paragraph as part of the edit.
Revision Principles
- Minimal changes: Only modify what is necessary to address the review comment. Do not rewrite entire paragraphs when a sentence-level edit suffices.
- Preserve structure: Do not reorganize sections or move paragraphs unless the review specifically requests it.
- No fabrication: Never invent data, results, or citations. If the review asks for additional analysis not present in the tables/figures, flag this to the user rather than making up results.
- Consistency: After each revision, ensure the changed text is consistent with surrounding paragraphs (transitions, terminology, notation).
- Scope discipline: Only address what the reviewer asked for. Do not make additional "improvements" beyond the scope of each comment.
Handling Special Cases
- Comment refers to a table/figure: Read the relevant table/figure
.texfile from the project folder before prop
Truncated for display — read the full file on GitHub.
Related Skills
x402
121.0kSet up Browser Use Cloud payments with x402 — pay per request from a crypto wallet (USDC on Base mainnet), no signup or API key.
algorithmic-art
177.9kCreating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems.
pptx
177.9kUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an em…
design
149.0kComprehensive design skill: brand identity, design tokens, UI styling, logo generation (55 styles, Gemini, Atlas Cloud, or MuAPI AI), corporate identity program (50 deliverables, CIP mockups), HTML presentations (Chart.js), banner design (22 styles, social/ads/web/print), icon design (15 styles, SVG…
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
