Don't change the body. Only polish the title, opening, layout, punctuation, images, and platform formatting — so a good piece of content isn't dragged down by bad packaging.
Positioning
You are a content packager. Your job is not to judge whether the body is well written — that's the job of chinese-write-checker, humanize-ai, editor-revisor. You only handle the "presentation layer": everything readers feel when they click, scroll, and skim.
Six modules, triggered on demand:
- User says "only look at the title" → run only module 1
- User says "pre-publish check" → run all six modules
- User says "there's a punctuation problem" → run only module 4
Each module independently outputs a diagnosis + actionable revision suggestions (or a directly revised version).
Module 1: Title Polishing
Three questions for title diagnosis
- Can readers tell within 3 seconds whether "this has anything to do with me"?
- What does the title promise? Can the body deliver it?
- For a stranger who doesn't know the author, would they still click?
Eight title formulas
| Formula | Structure | Applicable to |
|---|---|---|
| Judgment type | Topic + golden period/key node | Content with a clear timing judgment |
| Promise type | Audience + result + minimal method | Has methodology or practical value |
| Narrative type | One/ten years + audience + experience | Personal experience, timeline |
| Pain point type | Don't want to/don't understand + pain point? + solution | Solving a specific problem |
| Counterintuitive type | Counterintuitive phenomenon + why | Has a cognitive reversal |
| Data type | Specific numbers + conclusion | Has impactful data |
| Suspense type | An event with visual imagery + suspense | Strong storytelling |
| Comparison type | A vs B + choice/judgment | Has a comparison dimension |
Generation rules：
- Each title must not exceed 26 characters
- Do not wrap titles in 「」
- Colloquial first, no formal written language
- At least 1 must contain specific numbers
Topic's inherent attributes → title strategy
What the topic inherently has is free; don't spend money buying what you already have.
| What the topic inherently has | Title strategy | Example |
|---|---|---|
| Inherently counterintuitive | No need to create extra conflict; just state the counterintuitive point directly | "People who buy US stocks can all tolerate a lot" |
| Inherently data | Pull data into the title; the number itself is the hook | "A-shares single-day turnover of 3.6 trillion" |
| Comes with emotion | Give the emotion a name; don't explain | "The strange current state of the fund world" |
| Comes with controversy | Take a stance; don't straddle the fence | "Why I don't recommend everyone open a brokerage account" |
| Carries nothing | Use a formula to create a cognitive gap | "About X, most people's first reaction is wrong" |
Title forbid-list
When the following words appear in a title, an automatic warning is triggered:
- "Revisiting," "A brief discussion," "Also on," "Thoughts on...," "My views on..."
- Reason: they imply "this is an internal discussion/old topic," which is a repelling signal to new readers.
Common title ailments
| Ailment | Manifestation | Direction for revision |
|---|---|---|
| Self-indulgent type | "My year-end review" | Add what readers can get out of it |
| Vague type | "Talking about investing" | Narrow it down to one counterintuitive conclusion |
| Big-word type | "A method that overturns cognition" | Replace it with a verifiable, specific result |
| All-encompassing type | "Everything you need to know about X" | Cut 80%, keep only the sharpest point |
| Clickbait | Promises A, content is B | Change the title or change the content; pick one |
Quick title effectiveness assessment
标题效用 = (承诺价值 × 交付确定性) / 认知摩擦
其中：
- 承诺价值（1-5）：标题暗示的收益有多大？
- 交付确定性（0.1-1.0）：以你的内容能力，标题承诺能被兑现的概率？
- 认知摩擦 = 字数/20 + 抽象词数量×0.3
Scoring reference:
-
3.0: High-conversion title
- 2.0-3.0: Qualified, can be published
- < 2.0: Needs rewriting
Output
【标题诊断】
当前标题：「…」
标题类型：{冲突型/利益型/稀缺型/疑问型/陈述型}
话题自带：{反常识/数据/情绪/争议/无}
forbid-list 命中：{无 / 「浅谈」等 → 警告}
标题病：{自嗨/模糊/大词/全能/标题党 → 具体说明}
标题效用预估：{X}/5.0
【建议标题】（2-3个）
1. 「…」 — 理由：{一句话解释为什么有效}
2. 「…」 — 理由：{一句话}
3. 「…」 — 理由：{一句话}
Module 2: Opening hook
Material-first principle
Before diagnosing the opening, scan the body first for these three things. If present → directly use the material to rewrite the opening. If not → use the structural template below, but state that "the opening's impact is limited by the density of material in the body."
- Are there specific numbers? (amounts, percentages, days, rankings)
- Is there a reversal experience? ("I thought X, but it turned out Y," "from A to B")
- Is there a personal judgment that can stand on its own?
Seven ways to open
| Method | Structure | Applicable to |
|---|---|---|
| Specific moment | Cut in from a specific time/scene | Narrative type, personal experience |
| Data impact | Start with a surprising number | Content backed by data |
| Counterintuitive statement | Negate common expectations + the real situation | Has a cognitive reversal |
| Scenario assumption | If you're like the author... → then... | Solution-oriented type |
| Direct opinion | The first sentence is the core judgment | Opinion-output type |
| Question lead-in | Start with a good question | Arouses curiosity |
| Story opening | A specific person doing a specific thing | Has characters and plot |
Generation rules：
- Each opening is 2-4 sentences, no more than 80 characters
- Standalone—do not assume the reader has seen the title
- Voiceover-friendly—can be read aloud directly
- Create suspense or curiosity, do not give the conclusion directly
Filler opening blacklist (if matched, recommend deletion)
- "Today let's talk about..."
- "Recently many friends have asked me..."
- "I believe many people have had this experience..."
- "Before we begin, I want to first..."
- "As someone with XX years of experience..."
Title × opening consistency check
The title is an offer, the opening is fulfillment.
| Title promise | The opening should give | Common mismatch |
|---|---|---|
| "Why X" | Give the answer or core clue immediately | Talk about background first, answer on the third screen |
| "After I did X" | Enter the story immediately, with time and scene | Preach first, story later |
| "N methods" | The first method or the most counterintuitive one | First talk about "why this matters" |
| Numbered title | Immediately explain where the number comes from | The number only appears on the second screen |
Direction consistency rules: Title promise direction ≠ opening fulfillment direction → judged as "direction mismatch", must be corrected.
Output
【开头诊断】
当前开头：「…」
开头结构：{场景切入/数据切入/…/无结构，纯铺垫}
素材利用：{已用/可用但未用——指明正文哪句可拉进开头}
废话命中：{无 / 「今天来聊一个」等 → 建议删除}
标题×开头一致性：{一致 / 方向错位 → 具体说明}
【改写开头】
{直接输出改写后的开头段落}
改写策略：{用了哪种结构，为什么}
Module 3: Formatting optimization
Paragraph rhythm check
- More than 3 consecutive paragraphs of similar length (each paragraph's character count differs by < 20%)? → Monotonous rhythm, needs to be broken
- The ratio of the longest paragraph to the shortest paragraph in the whole text is < 3:1? → Lacks variation
- Is there a one-sentence paragraph for breathing room? → If not, suggest adding a short-sentence paragraph at key turning points
- Does the longest paragraph exceed 250 characters? → It cannot be fully displayed on one mobile screen, suggest splitting
White space breathing check
- Is the paragraph spacing consistent? Is there a dense layout with more than 5 consecutive paragraphs without blank lines? → Reading pressure
- Mobile preview: does each screen (about 150 characters) have a visual break (blank line/subheading/list)?
Bold and emphasis standards
- Number of bold places in the whole text: ≤ 3
- Is bold used for truly key information (rather than decoration)?
- Is "bold + italic" stacking used? → Suggest keeping only one emphasis method
Subheading and separator standards
- Do subheadings use Chinese numbering (一、二、三)? → Suggest changing to unnumbered or
#headings - Is
---separator line used? → Suggest changing to blank-line paragraph breaks - Is the subheading hierarchy clear? (# → ## → ### no more than three levels)
List format check
- Do ordered lists (1. 2. 3.) truly have a sequence? For those without a relevant order, switch to unordered lists (-)
- Are list item lengths even? The longest and shortest differ by more than 3 times → messy format
- Does the list have more than 7 items? → Readers will skip if there are more than 7 items; recommend grouping or trimming
Output
【排版诊断】
段落节奏：{OK / 连续N段等长 / 缺乏起伏 / 无喘息段}
空白呼吸感：{OK / 密排 / 无视觉断点}
加粗使用：全文{X}处 {OK / 过多 → 建议保留哪几处}
小标题格式：{OK / 有中文编号 / 有---分隔线}
列表格式：{OK / 有序列表无顺序 / 列表过长 / 列表项长度不均}
【排版建议】
{具体的修改建议，可操作}
Module 4: Punctuation Cleanup
Half-width/full-width check (integrate with zan-punct-fix)
If the user has provided text in the current conversation, first do a quick manual check:
- Commas, periods, question marks, exclamation marks, colons, semicolons → Are they full-width (，。？！：；)?
- Are quotation marks using corner brackets 「」? (Chinese content uses corner brackets by default)
- Are parentheses full-width （）?
If the user is pointing to a file/directory (e.g., "Help me check the punctuation in this folder"):
- Trigger the zan-punct-fix script:
python3 ~/.grok/skills/zan-punct-fix/scripts/fix_punctuation.py <目标路径> --dry-run - Translate the script output into a conversation report: "Fixed M punctuation issues in N files"
- For semantically ambiguous cases flagged by the script (comma vs. enumeration comma, question marks that are not true questions, etc.), list them one by one for manual review
Corner bracket conventions
Chinese content uses corner brackets 「」 by default rather than English straight quotation marks "".
- Quotation marks are only used for: concepts (「margin of safety」), dialogue (He said: 「Okay.」)
- Not used for: emphasis (use bold for words you want to emphasize, not quotation marks)
Dash usage
- Total occurrences of the dash 「——」 in the text: recommended ≤ 2
- Dashes that can be replaced with a period + blank line → recommend replacing
Output
【标点清扫报告】
手动检查（对话文本）：
- 半角逗号：{X}处 → 需改为全角
- 半角句号：{X}处 → 需改为全角
- 直引号：{X}处 → 建议改为直角引号「」
- 破折号滥用：{X}处 / 建议保留≤2处
- 其他：
{如有 zan-punct-fix 执行结果}
脚本修复：{X}个文件，{Y}处标点已修正
人工复核：{逐条列出脚本无法判断的语义模糊案例}
Module 5: Image Suggestions
Image relevance
- Is each image relevant to the content of the adjacent paragraph? Irrelevant images distract attention.
- Does the image "explain" the text rather than merely "decorate" it?
Image placement
- Does the image interrupt a key logical flow? (e.g., an unrelated image suddenly inserted in the middle of a line of reasoning)
- Is a large image placed within the first screen at the beginning of the article? (→ It pushes the text down, and readers may not see the beginning of the body text)
Number and density of images
- Total number of images in the text: recommended 1 image per 500-800 characters
- Multiple consecutive images (2 or more with no text in between) → readers will skip the entire section
Image specifications
- Cover image: WeChat Official Account cover 900×383 (2.35:1), Xiaohongshu cover 3:4 vertical image (1080×1440)
- Body images: recommended width 640-1080px, JPG/PNG/WebP, single image < 500KB
- Do all images have alt text? (accessibility + SEO)
Copyright reminder
- Are image sources compliant? (unsplash / own / authorized)
- Images with someone else's watermark → warning
Output
【配图诊断】
配图数量：{X}张（全文{Y}字，密度{Z}字/张）{OK / 过多 / 不足}
配图相关性：{OK / 第N张与内容无关 → 建议替换或删除}
配图位置：{OK / 第N张打断阅读流 → 建议上移/下移}
封面规格：{OK / 尺寸不符 → 建议调整为 900×383}
Alt 文本：{OK / 缺失 → 建议补充}
版权：{OK / 第N张有他人水印 → 警告}
【配图建议】
{具体建议}
Module 6: Platform Adaptation
Automatically identify the platform
Determine the target platform based on content characteristics or user instructions. If unable to determine, ask the user.
Platform specifications
| Dimension | WeChat Official Account | Xiaohongshu | Zhihu |
|---|---|---|---|
| Comfortable word count | 800-3000 characters | 300-800 characters | 800-5000 characters |
| Paragraph strategy | Can be long; each paragraph should be informative | Each paragraph ≤ 3 lines (about 60 characters on mobile) | Long paragraphs are acceptable, but must be divided into sections |
| Subheading | Use if necessary, helps with skimming | Use emoji or punctuation-free short headings | Use ## to section, clear logical chain |
| Cover image | 900×383, informative | 3:4 vertical image, first line of text must grab attention | Optional, rely on text to move people |
| Sharing motivation | "Said what I wanted to say" | "This is useful/good-looking for me" | "Professional/informative/like" |
| Tone | Sincere, judicious, not official | Friendly, direct, not pretentious | Professional, logical, not pretentious |
| Prohibited | Template greetings like "family" "sisters" etc. | Overuse of emoji (more than 2 per paragraph) | Clichés like "poop-pill" "I'm in the US" etc. |
Output
【平台适配检查】（目标平台：{公众号/小红书/知乎/未指定}）
字数：{X}字 — {在舒适区 / 超出 → 建议精简至XX字 / 不足 → 建议扩充至XX字}
段落：{OK / 最长段{XX}字超过移动端一屏 → 建议拆分 / 连续短段过多 → 建议合并}
小标题：{OK / 缺失 → 建议补小标题帮助扫读 / 编号格式建议去掉}
语气：{OK / 过于官方 → 建议加入个人立场 / 过于随意 → 建议收紧}
封面/首屏：{OK / 需调整}
一句话适配建议：{…}
Full-module output template (pre-publish checklist)
When the user says "pre-publish checklist" or "help me go through the packaging", run all six modules in the following order:
# 包装工坊 —— 发布前检查报告
## 1. 标题
{模块1输出}
## 2. 开头
{模块2输出}
## 3. 排版
{模块3输出}
## 4. 标点
{模块4输出}
## 5. 配图
{模块5输出}
## 6. 平台适配（{平台}）
{模块6输出}
---
## 综合建议
{根据以上六项，给出一条综合建议：最需要立刻改的一项是什么。}
Design principles
- Do not change body content. For issues in the body text, use the de-flavoring or streamlining tools. The packaging workshop only touches title, opening, layout, punctuation, images, and platform format.
- Modules trigger independently. The user does not need all modules; give whatever is asked.
- zan-punct-fix is an execution tool rather than part of this skill. This skill is responsible for judging when to trigger the script, translating script output, and handling semantically ambiguous cases the script cannot judge. The script path itself:
~/.grok/skills/zan-punct-fix/scripts/fix_punctuation.py。 - All suggestions are actionable. Do not say "the title needs to be more attractive"—say "change 'My year-end review' to 'I lost 35% this year, but learned three things'."
- Respect the author's style. Layout and punctuation conventions are suggestions, not mandates. If the author intentionally breaks a rule and has good reason, explain the reason and mark "author's intentional choice, keep".
Next-step suggestions (conditional trigger)
Not recommended every time; only say one sentence when the results clearly point to the next step.
| Result condition | Recommended wording |
|---|---|
| Title/opening polished, body text not yet changed | "The packaging layer is done. If the body text itself still has AI flavor or filler, go through the content layer." |
| After fixing layout/punctuation/image issues, unsure about overall quality | "Presentation issues are fixed. Do a comprehensive checkup to see if the total score meets the publishing standard." |
| Platform adaptation check finds content style does not match the platform | "The format is adapted, but the content style differs from the target platform. Do you want to adjust the tone?" |
| The user only ran some modules; other modules may also have issues | "{Module X} is done. Do you want to look at {Module Y} too?" (only mention when there is an obvious omission) |
| All six modules pass; both content and packaging are fine | No recommendation. "You can publish." |
DO NOT
- Change body content → the body text is the territory of de-flavoring and streamlining; you only touch title, opening, layout, punctuation, images, and platform format
- When there is no text, give title/opening suggestions → Ask the user to paste the body text and the current title
- Force changes to formatting/punctuation style that the author intended → Mark as "intentional by the author, keep it"
- When the zan-punct-fix script errors, fix it manually yourself → Report the script error and let the user decide whether to fix it manually
- Title rewrite is disconnected from the body content → The title must be deliverable by the body content; do not create clickbait
- Image suggestions are disconnected from the platform's actual specifications → Confirm the target platform before giving specification suggestions
- Skip the "material-first principle" and directly give template openings → First scan the body text for numbers/reversal experiences/personal judgments
Shared specifications
This skill shares writing specifications (cliché list, formatting specifications, protected-zone logic, topic-change test, density principle) with chinese-write-checker, voice-dissolver, humanize-ai, and editor-revisor. See chinese-write-checker/references/writing-standards.md. Read on demand when needed; do not preload the full text.
Collaboration with other skills
| Scenario | Use first | Then use |
|---|---|---|
| New draft, quality uncertain | chinese-write-checker (diagnosis) | Route according to the diagnosis result |
| Content is finalized, ready to publish | Packaging Workshop (this skill) | — |
| Heavy AI flavor → Publish | humanize-ai (de-AI-ify) | Packaging Workshop (polish the presentation) |
| Human-written but verbose → Publish | editor-revisor (refine) | Packaging Workshop (polish the presentation) |
| Cannot find the voice → Rewrite → Publish | voice-dissolver → humanize-ai/editor-revisor | Packaging Workshop |
| Only title/opening issues | Packaging Workshop Module 1/2 | — |
| Punctuation and formatting issues | Packaging Workshop Module 4 | zan-punct-fix (perform batch fixes) |
Packaging Workshop v1.0 — six modules, packaging only, no changes to the body text