Chinese Thesis Workbench Skill
Chinese Thesis Workbench Skill turns a real project — source code, database, screenshots, and literature — into a traceable, editable, school-template-matching Chinese undergraduate thesis DOCX, step by step.
Last reviewed by the paperbanana team on Jul 13, 2026
Install
git clone https://github.com/ZyhSechub/chinese-thesis-workbench-skill ~/.claude/skills/chinese-thesis-workbench
pip install -r requirements.txt
python scripts/workspace/init_thesis_workspace.py <your-thesis-project-dir>What is Chinese Thesis Workbench Skill?
Chinese Thesis Workbench Skill (chinese-thesis-workbench) is a Claude Code skill that turns a real project — its source code, database schema, screenshots, and literature — into a Chinese undergraduate graduation thesis, following the school's own template and the style of past sample papers. It is explicitly not a one-click paper generator: the README frames it as a "standardization workbench" that breaks the thesis into reviewable, reversible steps — collect materials → choose a delivery path → build evidence → confirm the outline → write chapter by chapter → generate/fill/edit the DOCX → run quality checks.
Its defining design choice is the four DOCX delivery paths, selected during intake and recorded in user-decisions.md: M1 generates a brand-new document in a built-in default style; M2 generates a new document but pulls high-confidence font/size/indent values from an analyzed sample paper; M3 copies a school template's DOCX shell and only replaces placeholder text and inserts chapter body content, leaving styles.xml, headers/footers, and section breaks untouched; and M4 edits an existing draft's paragraph text only, without touching paragraph or run properties. M3 is the recommended path for schools with strict formatting requirements, since it never attempts to recreate a template from scratch.
The skill enforces a hard evidence chain: it will not write formal thesis body text before materials are collected, standards resolved, and project evidence built, and it refuses to invent features, APIs, test results, or references. When source material is genuinely thin, a material-gap-handoff flow can — only with explicit user consent — search recent literature in the same field or run a brainstorming pass, with everything routed through content-decisions.md for the user to accept or discard. Chapter drafts get pre-set word budgets with automatic compression if exceeded, and a dedicated AIGC-style governance pass reduces vague, AI-sounding phrasing without adding new facts.
Core capabilities
Four DOCX delivery paths (M1-M4)
M1 default-style generation, M2 sample-style-approximation generation, M3 school-template-copy filling (preserves original OOXML shell, only replaces placeholders and inserts text), and M4 existing-draft text editing via anchor-based replace/insert/delete.
Evidence-driven governance layer
paper-context/ tracks workflow state, standards, project evidence, literature, AIGC reports, and Word-comment revisions; every major factual claim must trace to evidence or verified literature before it can appear in the thesis.
Sample-paper and template analysis
analyze_docx.py extracts chapter word counts, figure/table rhythm, paragraph styles from past sample papers, and page setup / style definitions from school templates, feeding both the outline and the chosen delivery path.
Literature governance pipeline
A fixed sequence — build reference pool → verify → filter → format → generate a verification checklist — extracts citations from literature PDFs and cross-references them against the thesis.
Word-comment-driven revision
Extracts an advisor's Word comments, revises point by point, and logs the change history without discarding the original document formatting.
User-facing workflow dashboard
Five control files (user-dashboard.md, material-inventory.md, content-decisions.md, blocker-report.md, user-decisions.md) surface current progress, missing-material impact, pending decisions, and blockers at every stage.
What you can use it for
Writing a thesis from an existing capstone project
Feed in your software project's source code, database schema, API list, and test results; build_evidence extracts project facts that back Chapter 4 implementation claims, including binding an E-R diagram to system-design chapters.
Matching a strict school-template format
Use the M3 path — apply_textual_edits.py --from-template school_template.docx — to fill your thesis into a copy of the template's DOCX shell without disturbing its styles, headers/footers, or section breaks.
Approximating a strong sample paper's look
Analyze a past excellent thesis with analyze_docx.py, then pass --sample-analysis to generate_thesis_docx.py so high-confidence style values (font, size, indent) merge into the default profile.
Revising after advisor comments
Extract Word comments from a returned draft and use the M4 edit path to make targeted paragraph-text revisions and record a change log, without rebuilding the document's styles.
Handling thin source material
When required materials are missing or the literature pool is empty, trigger the material-gap-handoff flow to search recent same-field literature or brainstorm writing angles — only with explicit user approval for each addition.
How to get started
- 1
Install dependencies and initialize the workspace
Run pip install -r requirements.txt, then python scripts/workspace/init_thesis_workspace.py <your-thesis-project-dir> to scaffold thesis-ai-standard/ and paper-context/.
- 2
Configure the standard profile
Edit <project-dir>/thesis-ai-standard/templates/standard-profile.yaml to reflect your school's formatting rules, then load the skill in Claude Code and ask it to organize the project into a thesis.
- 3
Analyze template or sample paper (M2/M3 only)
Run python scripts/docx/analyze_docx.py <template-or-sample>.docx --json-out result.json to extract style and structure signals before generation.
- 4
Generate or fill the DOCX and run quality gates
Depending on the chosen path, run generate_thesis_docx.py (M1/M2) or apply_textual_edits.py --from-template (M3) or apply_textual_edits.py with --replace/--insert/--delete (M4), then let the workflow run its standards, evidence, reference, figure, and AIGC quality checks before delivery.
How it compares to similar skills
Chinese Thesis Workbench Skill is narrowly built for the Chinese undergraduate 毕业论文 format and DOCX delivery constraints that other academic-writing skills in this directory do not address.
Academic Research Skills
Pick Academic Research Skills for an English-language, discipline-agnostic research → write → review pipeline aimed at journal or conference submission rather than a Chinese school-template thesis.
PaperOrchestra
Pick PaperOrchestra if you need a LaTeX conference-paper pipeline built around a published multi-agent benchmark, rather than DOCX thesis delivery matched to a specific school format.
AI Research Feedback
Pick AI-research-feedback if you need referee-style critique of an already-written LaTeX paper rather than end-to-end Chinese thesis drafting and formatting.