doc-normalize

name: doc-normalize description: Scan a Word document for structural normalization issues and return a report. Use when the user asks to normalize headings, numbering, outline structure, or style consistency, or when they invoke /doc-normalize.

DOCX normalization scan

Goal

This skill performs document normalization analysis for .docx targets.

V1 is scan-only. It must not modify the target document.

When To Use

Use this skill when the user wants to:

• inspect a Word document for pseudo headings or pseudo lists
• assess whether a document follows heading and numbering semantics
• check style drift against an existing template or stable target document
• run /doc-normalize

Do not use this skill to silently normalize a document during ordinary drafting. Ordinary drafting should preserve the current local block and inherit the target document's existing semantics.

Workflow

1. Resolve the target .docx and any authoritative template or stable target document.
2. Use the docx skill to inspect the document structure and any relevant OOXML or faithful extraction output.
3. Capture a style baseline:
   • heading styles or heading-like paragraphs
   • numbering conventions
   • obvious paragraph spacing and table conventions if relevant
4. Scan for normalization findings.
5. Return a report. Do not mutate the document in v1.

Scan Focus

Prioritize findings that are structural and high-confidence:

• pseudo headings such as plain paragraphs that visually look like 1., 1.1, 一、, or similar section markers
• pseudo lists such as manually typed 1., (1), -, or bullet-like prefixes where true numbering should be used
• paragraph style drift where nearby equivalent blocks use a stable style system
• heading hierarchy mismatches that are obvious from surrounding structure

Lower-confidence items may still be reported, but must be marked as review-first:

• TOC or cross-reference inconsistencies
• large-scale template drift
• complex table formatting anomalies
• header, footer, image anchor, comment, or tracked-change issues

Output

Return a concise report with these sections:

1. Target document
2. Scan scope
3. High-confidence findings
4. Review-first findings
5. Risks and unknowns
6. Suggested next natural-language follow-ups

Suggested follow-ups should stay scan-oriented in v1, for example:

• "只继续分析标题和编号"
• "把高置信问题按章节列出来"
• "对照模板再检查一次样式漂移"

Guardrails

• Scan-only.
• Do not modify the target .docx in v1.
• Do not promise automatic normalization.
• If the target document or authoritative template is unclear, ask one blocking question.
• Prefer reporting uncertainty over inventing a normalization rule that is not clearly supported by the document.