Convert a complete mathematical modeling paper into an English conference manuscript with traceable evidence and a clear research storyline. Use the built-in IEEE template by default; do not treat "applying a template" or "English translation" as completing the conversion, and do not promise acceptance or EI indexing.
Inputs and Default Choices
The only required input: a complete mathematical modeling paper, which may be a PDF, Word, or LaTeX project. It includes the main text, equations, results, figures and tables, and references; missing attachments do not automatically mean the paper is incomplete. If only a title, abstract, or outline is available, first require the complete paper, and do not fabricate research from scratch.
Optional: code and data, original figures, target conference and call-for-papers link, page/language requirements, author affiliations, anonymity requirements. First check the paths and attachments provided by the user to avoid repeatedly requesting information already available.
- No target conference: continue producing a general English IEEE conference draft, record "target conference to be determined," and do not assume 6 pages or a specified number of figures.
- Only the paper: rewrite normally; note that numerical values come from the original manuscript, and do not claim reproduction. If PDF extraction fails, try available layout reading/OCR, and key equations and tables must be checked against the original pages.
- Code/data available: select necessary verification or recalculation according to the main conclusions; do not expand into a large-scale new experimental project unless requested. Conclusions lacking evidence should be downgraded, deleted, or listed as pending verification, and must not be padded with fabricated numerical values.
- Missing author information: use explicit author placeholders and list them as to-dos; handle anonymous review according to conference requirements, and do not fill in authors, affiliations, or funding out of thin air.
- Target conference has its own template: prioritize the target template provided by the user, while retaining this skill's rewriting process. IEEE is not the universal format for all EI conferences.
Workflow
1. Read through and determine the research storyline
Read the original manuscript, project description, and applicable AGENTS.md, and review existing results and figures/tables. Identify the actual research field, and do not guess the topic based on numbers such as "Competition Problem A/C."
Select the research question in the original manuscript that has the strongest evidence and can form a coherent contribution. Multiple competition sub-questions may be merged, or only the necessary parts may be retained; do not copy "Problem 1, Problem 2, Problem 3" one by one. In the revision notes, clearly state the research question, methods, supportable contributions, and scope. A combination of conventional algorithms cannot be directly packaged as an original creation.
2. Establish an evidence and figure/table inventory
Before rewriting, read the Rewriting and Verification Guide. Save in the output directory 改稿说明.md, and update it as work progresses:
- Mapping from original sections to conference manuscript sections and reasons for important deletions or changes;
- Core conclusions, original locations, numerical sources, and verification status;
- Where original figures/tables go in the new manuscript, and reasons for deletion, merging, or redrawing;
- Data nature, experimental boundaries, missing information, and pre-submission to-dos.
Distinguish real measurements, official data, simulation data, synthetic data, and schematic diagrams. Do not replace the original manuscript's assumptions with "real data," and do not write simulator results as hardware measurements.
3. Establish an independent LaTeX output project
Keep the original manuscript; by default, create a new one next to or within the paper project conference_ieee/. If a converted manuscript already exists, continue modifying the user-specified version without overwriting other projects.
First-time creation can be executed (SKILL_DIR represents the current skill directory; use the actual absolute path):
python3 "SKILL_DIR/scripts/init_manuscript.py" --output "/absolute/path/conference_ieee"
The script only copies the built-in template and creates the figures and verification directories; it refuses to overwrite existing directories.The initialized main.tex is a template example; its sample body text, authors, references, and sample images must be replaced before it can become a paper. For the template source, files, and scope of application, see Template Description。
4. Rewrite the paper and organize figures and tables
Organize the English manuscript around the problem, method, and evidence. It usually includes Introduction, Related Work, Problem Formulation, Method, Experiments and Results, and Conclusion; merge or adjust according to the actual research, and it is not mandatory for every manuscript to be exactly the same.
Retain variable definitions, constraints, algorithm details, parameters, and experimental settings that affect reproducibility. Numbers in the abstract must be consistent with the results tables; conclusions must not go beyond the experiments. Remove restatements of competition problem statements, organizational slogans, and answer content unrelated to the main line.
Choose figures and tables according to their argumentative purpose: method mechanism, main results, explanation of causes, and scope of applicability. Do not delete necessary evidence just to fill page counts or figure counts. When redrawing, prioritize reusing real data and executable plotting code; when only the original image is available, keep a clear and readable version or faithfully redraw it, and do not guess precise data. Each figure has an English caption, in-text citation, units, and a traceable source. Use conventional plotting tools for scientific result figures, not generative images to simulate curves or experimental results.
If literature needs to be added or citations need to be verified, search original papers, publisher pages, or official conference pages; do not fabricate DOIs, page numbers, conference names, or references. When the user requests conference selection or judgment of EI indexing, separately verify the current official information and explain the scope that has been verified.
5. Compile, check, and deliver
Keep the template's font size, column width, and margins; page limits are based on explicit conference requirements or user requirements. When over the page limit, prioritize changing content and figure/table layout, and do not conceal it with extreme font shrinking or negative spacing.
In a supported environment, open main.tex to the built-in LaTeX editor and invoke the compilation tool, fixing according to its limitations. When multiple files or external figures are not supported, keep the editor and source code, and use an existing full project compiler for verification; do not treat opening a preview as successful compilation, and do not install TeX for the built-in editor.
When local tools are available, you can use the following in the output directory:
latexmk -pdf -interaction=nonstopmode -halt-on-error -file-line-error main.tex
Adjust the command according to the actual compiler/font requirements. Check citations, cross-references, formulas, all key numerical values, and font/boundary warnings, and inspect the PDF page by page for figure clarity, cropping, float placement, and whitespace. When actual compilation is impossible, deliver the source code and clearly state the reason; do not claim that the PDF has been verified. When a full compiler is available, unpack and compile the final source package once in an independent directory to confirm that there are no absolute path dependencies or missing figures.
Usually deliver:
main.tex, necessary class files, references, and figures;- The successfully compiled
main.pdf； 改稿说明.md, including figure/table tradeoffs, evidence status, and unfinished submission information;- Necessary plotting scripts/data and a brief verification record in
validation/; - A source ZIP that can be compiled independently, without mixing in original sensitive data, useless caches, or machine-specific environments.
Completing revisions and local delivery require no additional approval; do not stop work that can be completed due to a lack of non-essential information. This skill does not include submission, payment of page fees, or contacting authors/conferences.