Imported from JustNSim/cn-thesis-docx-skills (skills/thesis-proposal-report-builder/SKILL.md). Install upstream with npx skills add JustNSim/cn-thesis-docx-skills --skill thesis-proposal-report-builder. Copyright stays with the author.

Thesis Proposal Report Builder

Core Use

Use this skill to turn a user's research introduction document into a Chinese university thesis/design proposal or opening report, with Markdown drafting and template-faithful DOCX output.

The proposal should emphasize:

• research background and significance;
• domestic and international research status;
• research objectives and contents;
• technical route and methods;
• feasibility analysis;
• schedule, expected outcomes, and references.

Required Decisions

Ask only when the user has not already specified a high-impact choice:

• whether to use the user's DOCX template or the bundled base template;
• whether the user needs only Markdown, only DOCX, or both;
• degree level: 本科, 硕士, 博士, or 其他;
• research title, research object, research-content boundaries, and contribution ownership;
• timeline assumptions and real milestone dates;
• citation style and whether numbering must follow first appearance.

Proceed with reasonable defaults for minor typography and file naming.

DOCX Preflight Gate

Before drafting Markdown or creating DOCX, read references/preflight.md and resolve its required fields. Do not draft Markdown until degree level is recorded. Do not create a DOCX until template mode, word-count target, and degree level are recorded.

If the user explicitly requests the bundled base template or supplies a template path, acknowledge that choice and do not ask it again. If template mode, word-count target, or degree level is missing, ask one concise preflight question before continuing.

Workflow

1. Read inputs

   • Research introduction, requirements, previous drafts, personal papers, references, figures, and optional template.
   • If using the bundled template, prefer assets/base-templates/proposal-template.docx.

2. Confirm proposal frame

   • Resolve the title, research object, problem definition, research contents, and expected contribution boundaries.
   • Keep the proposal distinct from the literature review: summarize related work only as needed to support problem definition and feasibility.

3. Plan the outline

   • Use the user's template if supplied.
   • Otherwise use a structure like: background/significance; domestic/foreign status; objectives and contents; technical route; feasibility; schedule; expected outcomes; references.
   • Use clear subsection titles for each research content and method.
   • Keep research objectives, research contents, and research schemes/methods one-to-one in number, order, and wording unless the user explicitly requests a different structure.
   • When the template provides heading numbering, write semantic heading text only. Do not prepend manual Chinese or Arabic numbers such as 一、, 1., or 2.1; let the template's numbering definitions render them.

4. Draft Markdown first

   • Produce outputs/proposal.md unless the user requests another path.
   • Add figure placeholders for research framework, technical route, data/evaluation scheme, and schedule if useful.
   • Mark unverified claims as TODO: 待核验.
   • Avoid formulaic AI-sounding patterns such as 不是...而是..., long dashes, and short parenthetical explanations when a normal sentence can explain the term.
   • Remove drafting prompts and template residue such as 研究目标与研究内容、关键科学问题, （下面分别介绍每个研究内容）, 请在此处, 模板示例, and unresolved TODO: notes before finalizing.
   • Follow references/numbering-style.md when body paragraphs need non-heading numbered points.
   • Run audit_markdown_report.py <draft.md> --degree <本科|硕士|博士|其他> --expect-research-items <N> --strict when the proposal has a known number of research contents; revise until degree alignment, one-to-one research-item alignment, citations, references, placeholders, and style findings pass.

5. Finalize references

   • Verify references when external lookup is available or when the user requests it.
   • Do not fabricate authors, years, venues, page numbers, DOI, URLs, or experimental claims.
   • Deduplicate by normalized title, DOI, arXiv ID, or URL.
   • Number by first appearance unless the required style says otherwise.
   • Every numbered reference must be cited in the body. Add a semantically relevant citation or remove the unused reference.

6. Build and validate DOCX

   • Work on a copy of the template.
   • Edit mapped template slots in place. Never delete the template body and rebuild it with generic Heading 1, Heading 2, Title, Normal, or list styles.
   • Put the thesis title into the cover's existing large title position. Preserve that paragraph, table-cell, or content-control style; do not place the title only as a normal body heading.
   • Remove unrelated sample body content from the template, including sample images, captions, example tables, and placeholder paragraphs. Keep only fixed template chrome such as cover labels, TOC, headers, footers, and required structural pages.
   • Preserve the cover, TOC, section breaks, headers, footers, page-number fields, and template styles unless the user explicitly requests their removal.
   • Do not apply blanket run-level font or size overrides. Use the template's existing paragraph and character styles.
   • Preserve template styles, section settings, headers/footers, TOC fields, and heading levels.
   • When using a DOCX template, do not rebuild the table of contents as plain paragraphs or manually-created hyperlinks. Keep the original Word TOC field envelope (w:fldChar begin/separate/end plus w:instrText) and replace only field results when unavoidable. If Word or LibreOffice cannot update fields, mark the TOC field dirty and disclose that page numbers require F9 update in Word.
   • Preserve template heading numbering. If Heading 1 or Heading 2 numbering disappears, run preserve_template_numbering.py <output.docx> --template <template.docx> --strict, then verify Heading 2 resolves to numbering level ilvl=1.
   • Keep bibliography numbering independent from heading numbering. References should render as bracketed [1], [2], ... and must not share the heading numId.
   • Convert citations to Word REF fields when renumbering may be needed.
   • Keep in-text citations as superscript Word REF fields, not plain baseline text.
   • Keep reference-list hanging indent compact so continuation lines align near the text after [n].
   • Update all fields and the TOC in Word or LibreOffice after headings and page breaks change.
   • Run inspect_docx_template.py <output> --strict and audit_docx_report.py <output> --title "<论文题目>" --strict before delivery.
   • Run audit_docx_structure.py <output> --template <template.docx> --strict before delivery when a DOCX template is used.
   • Run compare_md_docx_headings.py <draft.md> <output.docx> --strict after DOCX conversion and fix any heading-name or heading-level drift.
   • Read the audit metrics, not only its exit code. Require a valid large cover title, at least one superscript REF field when references exist, no plain citations, complete bibliography/bookmark and field/bookmark mappings, no uncited or missing references, no duplicated figure/table-list entries, and zero indent, duplicate-reference, or field-error findings. See references/quality-gates.md for the exact checklist.
   • Render and inspect pages. If field update or visual inspection is unavailable, report that limitation instead of claiming layout validation.

Figure And Table Guidance

• When the user asks for figure support, provide the figure title, recommended placement, key elements, and an optional drawing prompt. Do not generate images unless the current agent environment explicitly supports image generation and the user requests it.
• Use concise, formal noun-phrase captions for tables and figures. Avoid sentence-style captions such as "介绍了...", "展示了...", "说明了...", or "分析了...".
• Prefer captions like "研究内容与技术路线", "研究计划安排", or "实验数据与评价指标".

References To Load

• Read references/preflight.md before drafting or creating DOCX.
• Read references/quality-gates.md before delivering Markdown or DOCX.
• Read references/docx-crossrefs.md before modifying numbered references, Word fields, superscript citations, or bibliography order.
• Read references/template-policy.md before using, anonymizing, or publishing bundled templates.
• Read references/numbering-style.md before drafting or revising body text that uses non-heading numbered points.

Bundled Scripts

• scripts/inspect_docx_template.py: audit template metadata, comments, tracked changes, hidden text, sensitive text hits, and external relationships.
• scripts/privacy_scrub_template.py: scrub DOCX templates before public release.
• scripts/audit_markdown_report.py: inspect Markdown drafts for degree alignment, citation/reference coverage, and common AI-sounding wording patterns.
• scripts/audit_docx_report.py: inspect bibliography paragraphs, citation fields, superscript state, duplicate brackets, and bookmarks.
• scripts/audit_docx_structure.py: inspect template fidelity for Heading 1/2 numbering, real updateable TOC fields, figure-list page breaks, and independent bracketed bibliography numbering.
• scripts/preserve_template_numbering.py: copy heading numbering styles and definitions from the template, then assign direct Heading 1/2 numPr without overwriting bibliography numbering.
• scripts/compare_md_docx_headings.py: compare Markdown heading names and levels with DOCX heading paragraphs after conversion.
• scripts/clear_update_fields_on_open.py: remove Word's update-on-open setting after fields have already been updated and saved.
• scripts/convert_refs_to_crossrefs.py: convert plain numeric citations and bibliography entries into Word REF fields and automatic [n] reference numbering.

The converter accepts numeric citations anywhere inside a simple text run, including forms such as GenProg[1]以来, [1,2], [1-3], and [1][2][3]. Combined and range citations are expanded into adjacent superscript REF fields, e.g. [1][2][3]. It intentionally rejects citations in complex rich-text structures, missing-reference citations, and oversized ranges so that it never silently loses formatting.

Install Python script dependencies when needed:

python -m pip install -r scripts/requirements.txt

Output Defaults

• outputs/proposal.md
• outputs/proposal.docx

Chinese-only projects may use outputs/开题报告.md and outputs/开题报告.docx when that better matches the user's existing files.