word-check

name	word-check
description	Document quality checker for academic papers. Performs cross-reference validation (figures, tables, equations) and format compliance verification against user-provided requirements. Reports issues with precise locations and severity levels without modifying the document. Triggers: 检查引用, 交叉引用, 检查格式, 格式验证, 核对, 是否符合要求, check references, cross-reference, verify format, compliance.
allowed-tools	Read Bash Glob Grep mcp__word-document-server__get_document_text mcp__word-document-server__get_document_outline mcp__word-document-server__find_text_in_document mcp__word-document-server__get_document_info mcp__word-document-server__validate_document_footnotes mcp__word-document-server__get_all_comments mcp__word-document-server__get_paragraph_text_from_document mcp__word-document-server__get_document_xml mcp__docx-mcp__validate_paraids mcp__docx-mcp__audit_document mcp__word-mcp-live__diagnose_layout
metadata	{"version":"1.1.0","category":"verification","upstream-skills":["word-read","word-format","word-edit"],"downstream-skills":["word-format","word-submit"]}

Word Checker

Overview

Word Checker is the quality gate for academic paper Word documents. It performs two types of checks — cross-reference validation and format compliance verification — and produces structured reports. It is read-only: it finds and reports problems but never modifies the document.

When to Use

• After formatting (word-format) to verify all changes applied correctly
• After content editing (word-edit) to verify cross-references still valid
• Before submission (word-submit) as a final quality gate
• Any time the user wants to validate their document

When NOT to Use

• To fix issues → use word-format or word-edit
• To analyze document structure → use word-read

Mode Detection

IF user mentions "交叉引用/cross-reference/检查引用/检查图表"
  → Cross-Reference Mode (Mode A)

ELIF user mentions "格式/format/compliance/是否符合要求/检查格式"
  → Format Compliance Mode (Mode B)

ELIF user mentions "结构/structural/paraId/bookmarks/文档结构"
  → Structural Validation Mode (Mode C)

ELIF user mentions "全面检查/full check/检查一下"
  → All Modes (A + B + C)

DEFAULT → Both Modes (A + B)

---

Mode A: Cross-Reference Check

Prerequisites

• Document Map from word-reader (.word-agent/{name}.map.md sidecar file, or conversation context)
• The extraction scripts below read the full text directly from the docx file inside Python — the full text never enters the conversation context

超大文档降级：对百页级文档（如学位论文），若一次性扫描内存/耗时过大，按 Document Map 的章节段落范围分章读取并扫描，再合并各章的定义/引用集合后统一应用 R1-R7。

Step 1: Extract All Asset Definitions

Scan the document for figure/table/equation definitions (captions). 编号支持四种形态：普通编号（图1）、章节式编号（图2-1、图2.3）、补充材料编号（图S1 / Table S1）、附录编号（表A1）。

python3 << 'PYEOF'
import re
import json

FILE_PATH = "{file_path}"  # ← 填入 docx 绝对路径

# 直接从 docx 文件读取全文（不经过对话上下文）
try:
    from docx2python import docx2python
    with docx2python(FILE_PATH) as doc:
        text = doc.text
except ImportError:
    from docx import Document
    text = "\n".join(p.text for p in Document(FILE_PATH).paragraphs)

NUM = r'(\d+(?:[-.．]\d+)?)'   # 普通/章节式编号: 1, 2-1, 2.3, 2．3

def norm(n):
    return n.replace('．', '.')

def defs(patterns):
    out = []
    for pat in patterns:
        out += [norm(m) for m in re.findall(pat, text, re.MULTILINE)]
    return sorted(set(out))

result = {
    # 普通 + 章节式编号
    "figures_defined": defs([
        r'^[  ]*图\s*' + NUM + r'[\s.．:：]',
        r'^[  ]*(?:Figure|Fig\.)\s*' + NUM + r'[\s.:]',
    ]),
    "tables_defined": defs([
        r'^[  ]*表\s*' + NUM + r'[\s.．:：]',
        r'^[  ]*Table\s*' + NUM + r'[\s.:]',
    ]),
    # 公式编号（行末右对齐），支持章节式 式(3.2)
    "equations_defined": defs([
        r'[（(]\s*' + NUM + r'\s*[)）]\s*$',
    ]),
    # 补充材料编号 (图S1 / Figure S1 / Table S1 / 表S1)
    "figures_S_defined": defs([
        r'^[  ]*图\s*S(\d+)[\s.．:：]',
        r'^[  ]*(?:Figure|Fig\.)\s*S(\d+)[\s.:]',
    ]),
    "tables_S_defined": defs([
        r'^[  ]*表\s*S(\d+)[\s.．:：]',
        r'^[  ]*Table\s*S(\d+)[\s.:]',
    ]),
    # 附录编号 (表A1 / 图A1)
    "figures_A_defined": defs([r'^[  ]*图\s*A(\d+)[\s.．:：]']),
    "tables_A_defined": defs([r'^[  ]*表\s*A(\d+)[\s.．:：]']),
}
print(json.dumps(result, ensure_ascii=False))
PYEOF

Step 2: Extract All References

Scan for in-text references（同样直接从文件读取；编号命名空间与 Step 1 一一对应）:

python3 << 'PYEOF'
import re
import json

FILE_PATH = "{file_path}"  # ← 填入 docx 绝对路径

try:
    from docx2python import docx2python
    with docx2python(FILE_PATH) as doc:
        text = doc.text
except ImportError:
    from docx import Document
    text = "\n".join(p.text for p in Document(FILE_PATH).paragraphs)

NUM = r'(\d+(?:[-.．]\d+)?)'

def norm(n):
    return n.replace('．', '.')

def refs(patterns):
    out = []
    for pat in patterns:
        out += [(m.start(), norm(m.group(1))) for m in re.finditer(pat, text)]
    return out

# Figure references（普通/章节式 + 补充材料 S + 附录 A）
fig_refs   = refs([r'(?:如)?图\s*' + NUM, r'(?:Figure|Fig\.)\s*' + NUM])
fig_refs_s = refs([r'(?:如)?图\s*S(\d+)', r'(?:Figure|Fig\.)\s*S(\d+)'])
fig_refs_a = refs([r'(?:如)?图\s*A(\d+)'])

# Table references
tbl_refs   = refs([r'(?:如|见)?表\s*' + NUM, r'Table\s*' + NUM])
tbl_refs_s = refs([r'(?:如|见)?表\s*S(\d+)', r'Table\s*S(\d+)'])
tbl_refs_a = refs([r'(?:如|见)?表\s*A(\d+)'])

# Equation references（式(1)、式（3.2）、式 (2-1)、公式(1)、Eq. (1)、Equation (1)）
eq_refs = refs([r'(?:式|公式|Eq\.|Equation)\s*[（(]\s*' + NUM + r'\s*[)）]'])

# Format consistency
format_counts = {
    "fig_cn": len(re.findall(r'(?:如)?图\s*\d', text)),
    "fig_en": len(re.findall(r'(?:Figure|Fig\.)\s*\d', text)),
    "tbl_cn": len(re.findall(r'(?:如|见)?表\s*\d', text)),
    "tbl_en": len(re.findall(r'Table\s*\d', text)),
}

def pack(items):
    return [{"pos": p, "num": n} for p, n in items]

print(json.dumps({
    "figure_refs": pack(fig_refs), "figure_S_refs": pack(fig_refs_s), "figure_A_refs": pack(fig_refs_a),
    "table_refs": pack(tbl_refs), "table_S_refs": pack(tbl_refs_s), "table_A_refs": pack(tbl_refs_a),
    "equation_refs": pack(eq_refs),
    "format_counts": format_counts,
}, ensure_ascii=False))
PYEOF

注意：S（补充材料）与 A（附录）编号是独立命名空间，连续性检查（R2）应在各命名空间内分别进行；章节式编号（2-1, 2.3）的连续性按章内序号判断。

Step 3: Apply Check Rules

See ../../references/cross_ref_rules.md for the complete rule definitions.

Apply R1-R7 in order:

Rule	Check	Severity
R1	Reference target exists	❌ ERROR
R2	Numbering is sequential	❌ ERROR
R3	No duplicate numbers	❌ ERROR
R4	First-reference order	⚠ WARNING
R5	Format consistency	⚠ WARNING
R6	No orphan assets	⚠ WARNING
R7	Range reference validity	ℹ INFO

Step 4: Compile Report

## Cross-Reference Check Report: {filename}

### Figures
- Defined: {list}
- Referenced: {list}
- ✅/❌ {details per rule}

### Tables
- Defined: {list}
- Referenced: {list}
- ✅/❌ {details per rule}

### Equations
- Defined: {list}
- Referenced: {list}
- ✅/❌ {details per rule}

### Summary
- ❌ Errors: {n}
- ⚠ Warnings: {n}
- ℹ Info: {n}

---

Mode B: Format Compliance Check

Prerequisites

• Document Map from word-reader
• Format Spec (from word-format session) OR user-provided format requirements

Step 1: Obtain Format Spec

If Format Spec exists in conversation context → use it. If not → ask user for format requirements, then parse using ../../references/format_spec_parser.md.

Step 2: Check Each Rule

For each non-null rule in the Format Spec, verify the current document state:

Page Setup:
  - Read get_document_info for page size, margins
  - Compare against Format Spec

Fonts:
  - Sample paragraphs from each category (body, headings, captions, references)
  - Use get_paragraph_text_from_document or get_document_xml for formatting details
  - Compare font family, size, bold/italic against Format Spec

Spacing:
  - Check line spacing from document XML or formatting info
  - Check paragraph before/after spacing

Numbering:
  - Check heading numbering format from outline
  - Compare against Format Spec pattern

Headers/Footers:
  - Check presence and content from document info
  - Compare against Format Spec

Special Rules:
  - Check each special rule individually
  - E.g., "首行缩进2字符" → verify indent settings

Step 3: Compile Report

## Format Compliance Report: {filename}

### ✅ Compliant ({n}/{total})
- ✅ Page size: A4 ✓
- ✅ Margins: top=2.54cm bottom=2.54cm left=3.17cm right=3.17cm ✓
- ✅ Body font: 宋体/Times New Roman, 小四 ✓
- ...

### ❌ Non-Compliant ({n}/{total})
- ❌ Line spacing: required=1.5x, actual=2.0x
  → Affected: P90, P91, P95 (3 paragraphs)
- ❌ Caption font size: required=五号(10.5pt), actual=小四(12pt)
  → Affected: Fig 1 (P92), Fig 3 (P115)
- ...

### ⚠ Unable to Verify ({n}/{total})
- ⚠ Image resolution ≥300dpi: cannot check via current tools
- ...

### Summary
- Compliant: {n}/{total} rules
- Non-compliant: {n}/{total} rules
- Recommendation: {具体建议}

---

Mode C: Structural Validation

Prerequisites

• docx-mcp server available (if unavailable, skip with warning)

Step 1: Paragraph ID Validation

Call: mcp__docx-mcp__validate_paraids(file_path)
Check:
  - Zero duplicate paraIds
  - Zero missing paraIds
  - All paraIds are valid 8-digit hex format

Step 2: Comprehensive Structural Audit

Call: mcp__docx-mcp__audit_document(file_path)
Check:
  - No broken bookmarks (bookmarkStart without matching bookmarkEnd)
  - No orphan image references (blip pointing to missing rId)
  - No invalid numbering definitions (numId referencing non-existent abstractNum)
  - No style conflicts (duplicate styleId)
  - No broken hyperlinks
  - No missing content types
  - No relationship mismatches

Step 3: Compile Report

## Structural Validation Report: {filename}

### Paragraph IDs
- Total paragraphs: {n}
- ✅/❌ Uniqueness: {details}
- ✅/❌ Format: {details}

### Document Integrity
- ✅/❌ Bookmarks: {n} checked, {n} broken
- ✅/❌ Image references: {n} checked, {n} orphaned
- ✅/❌ Numbering definitions: {n} checked, {n} invalid
- ✅/❌ Style definitions: {n} checked, {n} conflicts
- ✅/❌ Hyperlinks: {n} checked, {n} broken

### Summary
- Errors: {n}
- Warnings: {n}

Fallback (docx-mcp unavailable)

⚠ docx-mcp 不可用，跳过结构验证（paraId、书签、图片引用完整性）。
  建议安装 docx-mcp 后重新运行检查。

Basic paraId duplicate scan via XML is available as minimal fallback. See ../../references/structural_validation.md.

---

Mode D: Layout Diagnostics (Live Mode Only)

Prerequisites

• Word must be open with the target document (live editing mode)
• word-mcp-live must be available

Trigger

"布局检查/layout check/排版诊断/页面溢出/orphan lines"

Step 1: Diagnose Layout

Call: mcp__word-mcp-live__diagnose_layout(file_path)
Returns:
  - Actual page count vs expected
  - Orphan/widow line locations
  - Text overflow areas
  - Unexpected page breaks
  - Heading/paragraph spacing anomalies

Step 2: Compile Report

## Layout Diagnostics Report: {filename}

### Page Analysis
- Total pages: {actual} (expected: {expected})
- Unexpected page breaks: {list}

### Typography Issues
- Orphan lines: {count} (locations: P{n}, P{m}, ...)
- Widow lines: {count} (locations: P{n}, P{m}, ...)
- Text overflow: {count} areas

### Recommendations
- {specific suggestions based on findings}

Fallback (Not in Live Mode)

⚠ 布局诊断仅在 Word 打开文档时可用（需要 Word 的排版引擎）。
  请在 Word 中打开文档后重新运行此检查。

---

Combined Mode (A + B + C)

When all modes are requested (or "全面检查"):

1. Run Mode A (cross-reference) first
2. Run Mode B (format compliance)
3. Run Mode C (structural validation) — skipped with warning if docx-mcp unavailable
4. Run Mode D (layout diagnostics) — only if live mode available, otherwise skipped silently
5. Merge into a single combined report
6. Provide overall recommendation

---

Post-Check Recommendations

Based on results, suggest next steps:

Result	Recommendation
All checks passed	"所有检查通过。可以使用 word-submit 准备投稿版本。"
Format issues only	"发现 {n} 处格式问题。使用 word-format 自动修复？"
Cross-ref issues only	"发现 {n} 处引用问题。需要手动修正后重新检查。"
Both types of issues	"发现 {n} 处格式问题和 {m} 处引用问题。建议先修正引用，再修复格式。"

Shared Resources

• ../../references/cross_ref_rules.md — Cross-reference check rule definitions
• ../../references/format_spec_parser.md — Format Spec structure
• ../../references/academic_formatting.md — Academic formatting knowledge
• ../../references/structural_validation.md — OOXML structural integrity rules (Mode C)
• ../../references/token_budget.md — Token efficiency rules