下载地址
https://www.zhengdianxue.com/wp-content/uploads/skills/org-vc3t81nj--plan-proofread-moways.zip将提示词发送给你的 AI 安装该 skills
请根据 https://skillhub.cn/install/skillhub.md,安装 @org-vc3t81nj/plan-proofread-moways。中文正式报告交付前审校校对——检查格式排版、文字标点、标题层级、图表一致性、数据溯源、AI 写作痕迹、模糊归因等问题,输出结构化修改建议。
阅读全文 (展开查看完整说明)
概述
plan-proofread: formal Chinese document proofreading
Use this skill as a professional proofreading workflow for formal Chinese reports. Focus on concrete problems, direct replacement suggestions, and delivery-risk items that would prevent the document from being submitted.
Safety boundary: this skill only analyzes user-provided documents and supporting materials. Bundled scripts are local read-only helpers; they do not send network requests, modify the source file, or access protected system locations.
Scope
- Enterprise development plans and strategic plans
- Feasibility study reports and bidding documents
- Board-ready or committee-ready formal reports
- Chinese long-form deliverables that need wording, formatting, punctuation, data, and table-of-contents review
Do not use this skill for fiction, poetry, scripts, or purely creative writing unless the user explicitly asks for formal proofreading only.
Bundled Resources
Scripts live in scripts/:
check_punctuation.py: scans common Chinese punctuation issues in.docxfiles.check_text_table_consistency.py: compares nearby text and tables for numeric consistency in.docxfiles.check_toc_and_pagination.py: checks table-of-contents entries, headings, chapter numbering, and page-number format in.docxfiles.check_data_sources.py: scans.docxfiles for data points without source attribution, vague attribution, and bare data; outputs "vague attribution (must fix)" and "suspected missing source (to confirm)" lists.
The scripts require python-docx. Run them only on user-provided .docx files, and treat their output as an initial screening that still needs human judgment.
References live in references/:
public-sector-compliance-review.md: public-sector and enterprise formal wording compliance review.punctuation-rules.md: Chinese punctuation rules and common AI-draft punctuation issues.toc-and-pagination-rules.md: table-of-contents and page-number failure patterns.text-table-consistency.md: text-table consistency patterns and a five-step checking method.error-categories-27.md: error taxonomy for tagging findings (30 categories including data-source types).publication-standards.md: quantitative error-rate and quality-rating reference.data-source-verification.md: data source provenance verification guide, including the "five-type source qualification" standard, vague attribution blacklist, and optional online verification rules.
Format Standards
Before proofreading, identify the applicable format standard:
| Standard | Use Case |
|---|---|
| CIECC-style consulting report standard | Reports explicitly following China International Engineering Consulting formats |
| Bowei V2.6 standard | Bowei consulting reports, or default when the user has no special requirement |
| User-provided standard | Any template, style guide, or client rule supplied by the user |
If standards conflict, follow the user-specified priority order. If no standard is specified, use Bowei V2.6 as the default.
Review Modes
| Mode | Use When | Steps |
|---|---|---|
| Full review | Whole document, long report, or user asks for comprehensive review | Run all steps |
| Lightweight review | Short excerpt or partial chapter | Skip framework scoring and full rating unless needed |
| Formatting-only | User asks only about layout or formatting | Focus on dimension 9 |
| Style-only | User asks to polish wording or reduce AI-like wording | Focus on dimension 5 |
| Text-table check | User asks about data/table consistency | Focus on dimension 6 and use the text-table script when possible |
Workflow
Step 1: Read the document
- Identify the input type:
.docx,.md,.txt,.pdf, or folder. - Read the content using an appropriate tool.
- If no file is available, ask the user to provide one.
For long documents, build a chapter index first. Review by chapter and maintain a short cross-chapter memo containing full-name/abbreviation pairs, data definitions, recurring indicators, and strategic-framework terms.
Step 2: Build a framework summary
For full review, output a short review frame:
- Document type
- Planning or reporting period
- Applicable format standard
- Core strategic framework and abbreviations
- Chapter index and table/figure count
- Approximate length and segmentation plan
- Main risk directions to inspect
Step 3: Review by priority
Review high-risk issues first, then language and formatting.
| Priority | Dimension | Focus |
|---|---|---|
| P0 | 0. Formal compliance wording | Official names, titles, institutional terms, current policy names, regional and external-facing wording, and user-provided authoritative wording requirements. Read public-sector-compliance-review.md when reviewing public-sector, state-owned enterprise, or external-release documents. |
| P0 | 3. Policy and legal references | Planning period, upper-level plans, policy document names, law names, article references, document-number format |
| P0 | 2. Logical consistency | Internal consistency, target-action alignment, framework alignment |
| P1 | 1. Basic text | Typos, missing words, grammar, punctuation, number format |
| P1 | 4. Terms and names | Full name/abbreviation consistency, professional terminology, first-use definitions |
| P1 | 6. Data and tables | Text-table consistency, data definitions, quantitative indicators, total rows, percentage totals, figure/table titles, figure/table numbering, data source provenance |
| P2 | 5. Formal style | Colloquial wording, overclaiming, repeated empty phrasing, AI-like wording |
| P3 | 7. Content completeness | Missing sections, missing implementation path, weak support |
| P3 | 8. Risk and compliance coverage | Missing risk categories, weak safeguards |
| P3 | 9. Layout and formatting | Format standard, headings, table-of-contents, page numbers, figure/table placement |
If a statement cannot be verified from the document or user-provided materials, mark it as 待确认 instead of inventing a correction.
Step 4: Text-table consistency check
For documents with tables or figures, check every table and nearby paragraph:
- Build a table/figure list with number, title, indicator, unit, period, totals, and percentage columns.
- Extract the nearby text window: five paragraphs before and two paragraphs after the table.
- Map numbers both ways: table to text and text to table.
- Check internal table consistency: totals, percentage sums, and obvious unit conflicts.
- Compare repeated indicators across chapters.
Use check_text_table_consistency.py as an initial screening for .docx files, then manually confirm each finding.
Step 4.5: Data source provenance check
For documents containing statistics, rankings, benchmarks, predictions, or "research shows…" claims, run a data source verification scan. Read references/data-source-verification.md for full rules.
The Iron Rule: Every data point (statistic, ratio, ranking, benchmark, prediction, "research shows…") must be traceable. Five qualification types — any one suffices: (1) named source, (2) footnote, (3) parenthetical citation, (4) scope/caliber declaration, (5) calculation reference. Bare data and vague attribution ("据相关数据显示""业内通行""网传") are always flagged.
Run the script:
C:UsersLenovo.workbuddybinariespythonenvsdefaultScriptspython.exe
~/.workbuddy/skills/plan-proofread/scripts/check_data_sources.py
report.docx --out data_source_report.mdThe script outputs two actionable lists:
- Vague attribution (必改): data points using blacklisted phrases — must add explicit source or downgrade to 【估算】
- Suspected missing source (待确认): bare data with no attribution — confirm each, then add source, mark 【估算】, or rewrite
Optional online verification: For a small number of key data points that (a) support core conclusions and (b) are publicly verifiable, use WebSearch to cross-check against National Bureau of Statistics, yearbooks, or listed-company announcements. Never search for client real operating data or unpublished data — confidentiality red line.
Step 5: Output the issue list
Output findings chapter by chapter. Sort by severity first.
Use this format:
【#序号】位置:章节/段落首句
原文:"……"
类型:XX | 程度:必改/建议优化/待确认
问题:……
改为:"……"For simple issues, a compressed line is acceptable:
【#序号】位置 | 类型 | "原文" -> "修改后" | 必改Allowed issue types:
错别字 · 漏字别字 · 标点错误 · 数字格式错误 · 语病 · 术语不规范 · 名称简称不统一 · 政策引用不准确 · 法律条文存疑 · 规划周期混淆 · 上下文逻辑不一致 · 表述过满 · 语言口语化 · AI写作痕迹 · 图表编号错误 · 图表标题位置不规范 · 页码目录不一致 · 排版格式问题 · 内容缺失 · 风险考虑不足 · 目标不量化 · 战略逻辑脱节 · 文表数据不一致 · 表内合计加和错 · 占比合计不等于100% · 单位量纲错配 · 口径同名不同义 · 改稿残留 · 合规用语待确认 · 数据缺来源 · 来源模糊 · 来源过时
Step 6: Output three summaries
After the detailed list, provide:
必改问题清单建议优化清单待确认清单
Step 7: Optional scoring and quality rating
For full review, use publication-standards.md as a reference to calculate:
- approximate checked characters
- number of mandatory corrections
- de-duplicated error count
- error rate per 10,000 characters
- quality rating and delivery suggestion
Quality ratings are advisory. Do not present them as a legal or regulatory judgment.
Step 8: Final self-check
Before delivering:
- Recheck cross-chapter consistency.
- Confirm high-risk dimensions were not skipped.
- Remove overconfident findings that should be
待确认. - Run text-table, punctuation, table-of-contents, and data-source scripts when the input is
.docxand the user asked for a full check. - Data source self-check: confirm no vague-attribution items remain unfixed; confirm all bare data points have been addressed (source added, marked 【估算】, or rewritten); confirm no client confidential data was searched online.
- Remind the user to refresh the table of contents and fields in Word before final submission.
Quick Rules
- Chapter numbering:
一、->(一)->1.->(1). - Use full-width punctuation in Chinese paragraphs.
- Use an em dash for year ranges, for example
2026—2030年. - Repeat the percent sign in ranges:
15%~30%, not15~30%. - Quote planning-period names consistently.
- Use six-corner brackets for official document numbers, for example
国发〔2025〕1号. - Figure titles go below figures; table titles go above tables.
- Do not add punctuation after attachment names.
- Prefer
其他; use截至as a preposition. - Every number cited in text should be checked against its source table or figure.
- Data source iron rule: Every data point must be traceable — named source, footnote, parenthetical citation, scope declaration, or calculation reference (any one suffices). Bare data and vague attribution ("据相关数据显示""业内通行""网传") must be flagged and either given a source or downgraded to 【估算】.
- Before submission, refresh the full table of contents and all fields in Word, then export PDF and inspect the table-of-contents pages.
Style Review Standards
Preferred style: formal, accurate, concise, conclusion-led, and stable.
Flag these patterns:
| Pattern | Examples | Severity |
|---|---|---|
| Colloquial wording | casual verbs, loose oral expressions | 建议优化 |
| AI-like wording | dense repeated transitions, empty paired slogans, repeated vague verbs | 建议优化 |
| Overclaiming | absolute claims without evidence | 必改 |
| Term misuse | wrong professional term or inconsistent abbreviation | 必改 |
| Reference issue | outdated or unverified policy/law reference | 必改/待确认 |
Execution Preferences
- Prioritize concrete delivery-risk issues.
- Use stable formal wording; do not over-polish into marketing language.
- Align the whole document with its stated strategic framework.
- Give replacement text whenever possible.
- Mark unverifiable items as
待确认.
Reference Index
| File | Content | When to Read |
|---|---|---|
public-sector-compliance-review.md | Formal compliance wording review for public-sector and enterprise reports | Public-sector, state-owned enterprise, or external-release documents |
punctuation-rules.md | Chinese punctuation rules and AI-draft punctuation issues | AI-drafted or heavily edited Chinese documents |
toc-and-pagination-rules.md | Table-of-contents and page-number checks | Before final delivery |
text-table-consistency.md | Text-table consistency failure patterns and checking method | Any document with tables or figures |
error-categories-27.md | Error taxonomy (30 categories including data-source types) | When tagging findings |
publication-standards.md | Error-rate and quality-rating reference | When producing full quality rating |
data-source-verification.md | Data source provenance verification guide: five-type qualification, vague attribution blacklist, online verification rules | Any document with statistics, rankings, predictions, or research citations |
Source Inspirations
Use these inspirations as general product-pattern references only. Do not cite them as evidence unless the current task requires verified sourcing.
| Source | Capability Pattern |
|---|---|
| WPS proofreading products | User dictionaries and formal document proofreading |
| Chinese writing assistants | Formal wording and typo screening |
| Office document proofreading APIs | Error taxonomy and structured findings |
pycorrector | Phonetic and shape-similar typo detection |
| Long-document proofreading workflows | Chapter-level segmentation |
| LanguageTool | Rule-based text checking |
| CY/T 266 | Error-rate and counting reference |
| GB/T 9704 | Official document formatting reference |