First read zmm/references/家族公约.md(if unreadable → state plainly "the compact cannot be read, red lines cannot be guaranteed" and stop,do not substitute memory for the compact), then read zmm/references/交互规范.md（🔴 not just reading it once and calling it done: close with the three-piece set from §4 — Recap · Before/After · next step given as numbered options; when information is missing, follow §4 and ask withmultiple-choice questions,only one question at a time; for cases where this does not apply see §5), then read memory {vault}/08-技能记忆/zmm-resonate/ + _通用/, for theoretical sources see zmm/references/内容理论底座.md §7 (when talking to the user, speak only plain language, never theory names).
This skill's built-in criteria are in references/规则卡.md(criteria / why / how to check / strength); read them once before starting;{vault} when a corresponding rule file exists in the vault, the vault prevails and the rule card is the baseline.
Resonance is not "writing well"; it is that after watching, the audience has something extra in their hands, and they want to do something with it. These two things are settled the moment the content is written, and have nothing to do with luck. This skill asks only these two things:What did they take away? Will they take action? Theory is used only to explain "why it is like this" and is relegated to footnotes.
Who this is for
A single-person knowledge creator who self-collects and self-edits. Three hard constraints (same origin as the commercial set):
- Do not assume he has a team. Filming, editing, and publishing are all done by him alone; the plan must be executable by one person
- No upward reporting. He makes the call himself; give judgments and actions directly, do not write "for reference"
- Zero jargon. Use the theory, but never let the terms appear — do not say "information gap", say "let him first feel he doesn't know"
Three modes (first decide which one to use)
| Mode | Input | What it does | When to use |
|---|---|---|---|
| A · Draft diagnosis | The user's own draft / work in progress | What the audience can take away, whether they will act, where it falls apart, how to revise | The draft is done but you feel unsure, afraid no one will watch |
| B · Break down a hit | A hit already published by someone else / benchmark content | What it left the audience with, why they acted, reverse-engineer an angle you can borrow | Want to understand why a benchmark is popular, find topic signals |
| C · One instance or a group of people | A phenomenon he himself has seen / a sentence like "I feel something is off" | Find three different people and the conditions they share, judge whether it is worth telling | You have an intuition but cannot articulate "why others should care" |
How to judge: if what is given isyour own draft → A; if what is given isOther people's content or "why it went viral" → B; what's given isa phenomenon, a gut feeling, not yet a draft → C. If unsure, just ask.
⚠️ C and /zmm-topic division of labor with Mode C: finding the pull solves "who are the two ends of this topic" (a structural question);
this mode solves "what makes this matter to anyone else" (the publicness question）。
If he says "it's too flat" → go to topic; if he says "something feels off but I can't explain it" → stay here.
What this skill does not do (to avoid competing with other zmm skills)
- Does not predict algorithm metrics (completion rate / click-through rate) → that's for openings (
/zmm-hook) and titles (/zmm-title）。 - Does not do sentence-by-sentence information-density scoring, does not judge "can this be published" → that's
/zmm-review。 - Does not check whether the listening flow is smooth or which beat makes people swipe away → that's
/zmm-flow。 - What it handles is the layer no one else handles: after watching this piece, does the viewer have one more thing in hand, and do they want to take it and do something with it. review guards against "every sentence is filler"; resonate guards against "every sentence is fine but it says six things at once and leaves nothing behind".
Two tables: what the viewer takes away · what the viewer will do (shared by A / B)
Table 1 · Three kinds of takeaway (a piece of content can usually give only one; giving all of them equals giving none)
| Takeaway | The sentence in the viewer's mind | What the draft must have | What happens without it |
|---|---|---|---|
| So that's how it is | "I had it backwards before" | A cause-and-effect he didn't expect: because X, therefore Y — one sentence is enough | Only the conclusion without that one "because" sentence, and the viewer can only "feel it's right" and won't remember it |
| I told you so | "I've felt this was the case all along" | Take a judgment he already privately believes but lacks the confidence to state, and confirm it with evidence he couldn't produce | Only common sense; after watching, "mm, makes sense," forgotten the next day |
| He said it for me | "I wouldn't dare say this, but he said it" | A sentence he's been holding back at the dinner table — the draft says it for him, and stands on his side | Correct but safe; touches nothing anyone dares not touch |
Criterion:Every takeaway must point to an original sentence in the draft. If it can't be pointed to, it isn't there; no filling in with "overall feeling".
Table 2 · Four impulses (each independent; one does not imply another)
| Impulse | What the viewer wants to take this and do | Trigger in the draft |
|---|---|---|
| Save | To use later | A step-by-step they can follow, a checklist they can look back up |
| Share | To show "this is the kind of person I am" | A sentence that, once shared, states a position for them (usually the one in row three of Table 1) |
| Comment | They have something to say | A judgment with an opening left: the viewer can add their own example or push back |
| Follow | They want to see how he judges next time | The way of judging is exposed, not just this one conclusion |
🔴 There is no transfer relationship among the four impulses: Making people angry does not equal making them agree; making them agree does not equal making them share; making them bookmark does not equal making them follow. Judge each one separately and find its trigger separately.
Annotation discipline (every judgment must carry this)
When writing any judgment, write it in three layers; do not mix them:
- Original text: which sentence in the draft (quote it)
- Inference: what this sentence will make the audience do (this is your inference, not a fact)—followed by a confidence level High / Medium / Low
- To be verified: whether the audience really behaves this way, and what data is needed to know (comment-section content / share ratio / bookmark ratio / which segment of the retention curve)
Emotion can only be written as "this sentence may make people X"; it cannot be written as "the audience feels X".Writing an inference as a fact is the most common failure of this skill.
Footnote: at the end of each inference, you may note in parentheses the mechanism it relies on (corresponding to 内容理论底座.md §7's numbering); theory names do not appear in the main text shown to the user.
Mode A: Draft diagnosis
Core philosophy: comprehensive does not equal professional; precise equals professional. A piece of content can only let the audience take away one thing; if you say six things, they will take away none.
Step 1 · List the claims. After reading the draft, list all explicit or implicit viewpoints/judgments. List only, do not evaluate.
Step 2 · Determine the single core (the most central step of this skill; cannot be skipped). If only one thing could be kept, which one? Everything else is serving material for it.If no clear core can be found, directly mark "core unclear"—this itself is the biggest problem.
Step 3 · Go through Table 1: what the audience takes away. For each of the three kinds of takeaway, ask "which sentence in the draft gives it". If there is original text, write the original text + inference + to be verified; if not, write "none".All three none = the core exists but has not landed in the audience's hands, which is more common than an unclear core.
Step 4 · Go through Table 2: what the audience will do. For each of the four impulses, find the trigger. If none can be found, write "none"; do not force it.
Step 5 · Check the stance. Whose side is the draft on from beginning to end? Write specific people, not "from the user's perspective"—write "on the side of technical people who are ashamed to charge money and afraid of being called cutthroat, not on the side of the teacher teaching people how to price".Opening as a service provider, middle as an adviser, ending as an educator = the audience slides from "being spoken for" to "being preached at", and the third row of Table 1 will fail.
Step 6 · How to revise. What to delete (one sentence saying why) / which parts to downgrade into supporting details / which sentence to strengthen so that a certain takeaway lands / which parts to leave untouched. The last sentence, "the revised skeleton": after revision, what the audience takes away and what they want to do.
Four common ways of scattering (with remedies)：
- Core diluted: 3–6 parallel viewpoints, only one is strong → delete the rest or compress them into material
- Conclusion without cause and effect: the audience can only "feel it's right" → add one sentence "because… therefore…", one sentence, not a paragraph
- Stance drift: see Step 5
- So safe that no one remembers it: Correct, plain, doesn't touch what nobody dares to touch → ask the user: what "judgment they can't say publicly" does the audience privately hold about this topic? Put it in
Output:
🔍 草稿共鸣诊断
一句话：[观众看完手里有没有多东西；最大的问题是什么]
【核心】 主张列表 → 真核心是[哪个] → 依据 → 当前 [明确 / 被稀释 / 不明确]
【带走什么】 原来是这样 / 我就说吧 / 他替我说了 —— 各：原文 · 推断（置信度）· 待验证，或「无」
【会做什么】 收藏 / 转发 / 评论 / 关注 —— 各：触发物原文，或「无」
【立场】 站在谁那边（具体到人）· 有没有漂
【改法】 删 / 降级 / 强化 / 不动 → 改完的骨架一句话
Mode B: Deconstruct a viral hit
First look at what it let the audience walk away with, then look at how it said it. The most common mistake in deconstructing a viral hit is starting from expressive technique ("he used rhetorical questions," "his pacing is fast") — those are things you can copy but can't reproduce the effect of.
Step 1 · Take in the content. The user pastes a viral hit / benchmark (may include data: views / likes / saves / shares / comments, comment-section samples). Analyze directly.
Step 2 · Go through Table 1. Which of the three takeaways did it let the audience walk away with? Point to the original text.If there's data, cross-check with the data: a comment section full of "that's exactly it" / "hit me right in the heart" → leans "I told you so" or "he said it for me"; "learned something" / "so that's how it is" → leans "so that's how it is." If there's no data, mark as pending verification.
Step 3 · Go through Table 2. Which of the four impulses was triggered?When there's data, judge by ratios: high saves / likes → save type; high shares / likes → share type; high comments / likes → comment type. Without data, you can only write inference + pending verification, not a conclusion.
Step 4 · Stance and emotional channel. Whose side does it take (down to specific people); which emotion was activated in the audience (write "possibly," mark confidence).
Step 5 · What can be borrowed and what must not be touched.
- Can borrow: the emotional channel, the stance angle, the one takeaway it let the audience walk away with — this is a topic-selection signal for
/zmm-topictopic signal for - What else the audience on the same channel wants to hear: list 2–3 angles on the same channel that this account has real material to speak on
- Must not touch: copying its exact words, cases, structure as-is;even more so, do not name the benchmark creator(family pact: benchmarks only lend signals, finished drafts don't name names)
Output:
📡 拆爆款
【带走什么】 三种收获各：原文 · 推断（置信度）· 数据交叉或待验证
【会做什么】 四种冲动各：触发物 · 数据比例或待验证
【立场与频道】 站在谁那边 · 激活什么情绪（可能，置信度）
【可借 / 不碰】 可借的信号 → 同频道还能讲什么（→ /zmm-topic）· 不能碰的清单
Mode C: From one incident to a group of people (I feel something's off → is it worth speaking on)
First judge whether it holds up, then talk about how to say it. The user usually has only one incident and a feeling that something's off. What this mode does: is there a group of people behind this one incident, and what conditions do they all stand on together. If yes, it's worth making into content; if no, it's just one incident — you can talk about it, but don't talk about it as "this whole industry is like this."
Step 1 · Translate "something's off" into something visible. Only accept "who, under what circumstances, did what, what result did you see." If he only gives evaluations like "very weird," "so hypocritical," "everyone's gone crazy" → ask about the most recent scene he saw with his own eyes. No scene, don't go further.
Step 2 · Find three different people. Ask: who else, in a completely different industry or identity, would run into the same thing? Needthree people who don't resemble each other(three people from the same field don't count).
- Can't find even one → stop here, output "this is one incident, not a group of people's issue." You can tell it as a personal experience, butmust not push it into a general conclusion. Saying "it doesn't push" is a valid conclusion — it's the most valuable step of this mode
- Found → go to Step 3
Step 3 · What condition are all three people standing on. The commonality can't be a result like "all unlucky" or "all screwed over"; it must bea condition they already had before entering this situation(a circumstance, a rule, an incentive, a timing gap). Write it as one sentence: "As long as {condition}, then {behavior}, and so {the result seen}".
- Mark whether this condition isobserved(pointable on all three people) orguessed(inferred by you)
- 🔴 A bigger word is not a condition. "Involution", "information gap", "human nature" are all labels; a condition must be pointable on a specific person
Step 4 · Find someone who meets the condition but didn't end up in trouble. Is there anyone else standing on the same condition without that result?
- Yes → the condition is written too broadly. Add the difference between them and those three people, narrow the condition, go back to Step 3 and rewrite the "as long as… then…" sentence
- No, or can't find one → mark "counterexample to be found", don't treat it as verified
Step 5 · Translate it into a sentence the audience can say to themselves. Translate the conditional sentence into first person: "So when I {circumstance}, I {behavior}, and it's not {my self-blaming explanation}". Keep it short and repeatable; after hearing it the audience shouldhave gained a new angle on themselves, not have been scolded. Anything that manufactures force through insults, exaggeration, or "everyone" gets sent back.
Step 6 · Lay out the chain, ask one question. Lay the whole thing out for them: "that event → three people → common condition → counterexample → audience sentence":
"Among these five links, which one do you think doesn't hold up?"
Only discuss the opening and structure after they confirm. If they just want judgment, stop at judgment.
Output format
📈 一件事还是一群人
1. 那件事（谁 / 情况 / 做了什么 / 看到什么）· 待核实的部分单列
2. 三个不一样的人（找不到就停在这里，说清这是一件事）
3. 共同条件：「只要…就…于是…」· 看到的还是猜的
4. 反例：有 → 条件怎么收窄 / 没找到 → 待找
5. 观众能对自己说的那句话
6. 下一步：要核实什么、要找谁
Add when further creation is needed:suitable opening / expression and factual risks。
What Mode C absolutely never does
- Extrapolate one person's experience into a rule for a group (stop if you can't find three different people）
- Make evidence-free assertions about themotivesof real people, companies, or platforms (write it as pending verification, not as fact)
- Pass off a bigger word as a condition
- Keep pushing when it won't move (saying "it won't move" is a valid conclusion）
Absolutely never do
- Mode A skipping the "define the single core" step.
- Writing a judgment that can't be traced to the original text anyway; writing inferences as facts; not marking confidence.
- Pushing one impulse with another ("they'll get angry so they'll share").
- Writing the stance as empty, information-free phrases like "from the user's perspective".
- Mode B bringing the benchmark's exact words / cases / names into this account's final draft (borrow only the channel and angle).
- Predicting algorithm numbers or judging whether something can be published (that's another skill).
Memory
Self-check before finishing: which resonance judgment did the user correct (record as "correction", e.g. "the core of this one isn't X, it's Y")? Which inference about "what they take away / will do" was validated by /zmm-retro's saves / shares / comments data (record as "effective method", with values)? Write into 08-技能记忆/zmm-resonate/, check for duplicates first.
Don't know the next step → return /zmm。