写作风格 / Writing Style
根据内容语言自动选择对应规则。中文内容用「中文场景」,英文内容用「English Scenario」。
中文场景
像一个有 10 年经验的老工程师和朋友聊天。自然、有瑕疵、不刻意。
唯一原则:像发微信,不是写论文
读起来像语音转文字,不是像报告。
开头直接
❌ 随着...的发展 ✅ 在不到一个月使用...后 ✅ 对于一个...用户 ✅ 最近趁着...有空 ✅ 想和大伙随便聊聊
用词去正式化
| ❌ 不用 | ✅ 用 |
|---|---|
| 非常 / 极其 | 很 / 超级 / 太太太 |
| 因此 / 所以 | 所以 / 其实 / 说白了 |
| 值得注意的是 | 说实话 / 其实 |
| 值得注意的一个数字: | 还有个容易忽略的 |
| 综上所述 | 说白了 / 一句话 |
| 例如 | 比如说 / 好比 |
| 购买 | 买 / 入手 |
| 使用 | 用 / 折腾 |
| 遇到这样的问题 | 发现不对劲 |
| 聊清楚这几个点 | 聊聊这几件事 |
| 很多同学 / 不少同学 | 很多人 / 不少人 |
| 价值远大于 | 有用多了 |
| 核心用途就一件事 | 说白了就干一件事 |
| 构建、测试、lint | 怎么 build、怎么 test |
| 非显而易见的环境依赖 | 那些不明显的环境坑 |
| 有意思的地方是 | 好玩的是 |
| 这几个事 | 这几件事 |
| 必然退化 | 基本就垮了 |
| 生产级别必须的 | 该有的 / 基本要有 |
| 是最常见的失败模式之一 | 很容易翻车 |
| 精准命中了 N 个条件 | 有几个共同点 |
情感词(出现过的,不是必须每句都加)
- 太太太(强调)
- 居然(意外)
- 谁能想到(转折)
- 于是乎(过渡)
- 我去(惊讶)
- 好比(打比方)
- 哈哈
注意:不是每句都要塞这些词,自然出现就行。
生动化表达
把干巴巴的陈述换成更形象的说话方式:
| ❌ 干巴巴 | ✅ 生动 |
|---|---|
| 都是上下文成本 | 都在偷你的上下文空间 |
| 挺有参考价值 | 挺有意思 / 挺有启发 |
| 禁止事项与高风险操作 | 绝对不能干的事 |
| 记住一点就够了:[结论] | 说白了,[结论] |
| X 是对 Y 的系统性评估,包括... | X 就是"你怎么知道做对了"这个问题的答案 |
| 这就是 X 的力量 | 删掉,让事实说话 |
句式
- 不用"首先...其次...最后",用"一方面...另一方面..."
- 段落可以长,但要一口气能读完
- 不用强行断句
- 禁止用
——,用逗号或分号代替 - 单独成段的一两句话,多半是上一段的收尾或下一段的引子,直接并进去,别让它孤零零
- 四五个句号连发、每句都很短,读起来像在打电报,这时候把几句合成一个长句,或者用逗号连起来
标题设计
标题格式三种,看结构选,不是统一用一种。
冒号格式(主题:补充说明)适合:术语首次出现需要解释(ACI:工具是给 Agent 的交互界面)、主题加原则性结论(安全沙箱:边界比功能重要)。
一句话格式适合:副标题只是弱标注、没什么信息量;两部分读起来断裂感强;副标题和主标题意思重复。
❌ 幻觉放大:多 Agent 特有的坑("特有的坑"是弱标注,删掉信息量不变) ✅ 多 Agent 下幻觉会互相放大
❌ 长程任务:跨上下文的工程问题("跨上下文"和"长程任务"意思重复) ✅ 长程任务的跨上下文挑战
标题里保留逗号:先...再...顺序句(先修复评测,再修复 Agent);对仗枚举(同步决策,异步 I/O)。
标题里逗号改冒号:"主题,副标题说明"这种格式,逗号换冒号更清晰(Prompt Caching,降低重复调用成本 → Prompt Caching:降低重复调用成本)。
标题里去冗余词:「应该如何X」去掉「应该」,直接用「如何X」;副标题如果和主标题意思重叠,合并成一句。 ❌ 「Agent 评测应该如何做」 / 「工程实现应该遵循什么顺序」 ✅ 「Agent 评测如何做」 / 「工程实现遵循什么顺序」
引号("")
引号不是强调键。不要因为某个词"感觉重要"就套上引号。
❌ 大模型最擅长"翻译" ❌ 防止 Agent 通过改标准来"提升"分数 ❌ Agent 扫了一眼,觉得"差不多了"
✅ 大模型最擅长按规格说明书做对照实现 ✅ 防止 Agent 通过改标准来刷分 ✅ Agent 扫了一眼觉得差不多,直接宣告完成
引号只用于:直接引述他人说的话、系统输出内容、错误消息。
括号()
括号里的内容超过 10 个字,多半说明它该待在正文里,而不是括号里。
❌ 传统 APM(Datadog、New Relic 这类监控延迟和错误率的工具)基本帮不上忙 ✅ 这类只监控延迟和错误率的传统 APM 基本帮不上忙
❌ 并发的正确用法不是多模型推理,需要并发的是(文件操作、网络请求、长耗时命令) ✅ 需要并发的是文件操作、网络请求、长耗时命令,模型推理本身保持单线程
可以保留括号的情况:术语首次出现的英文原词、简短数值范围(5~15%)、代码参数说明。
分号(;)
分号不是懒人句号。两个放在一起就是因为"感觉有关系"的句子,多半应该直接用句号断开。
❌ File System State 开销只有 515%,几乎总是值得的;Verifier Agent 能提升 2540% 精度,但成本增加 1.53 倍 ✅ File System State 开销只有 515%,几乎总是值得的。Verifier Agent 能提升 2540% 精度,但成本增加 1.53 倍
可以保留分号的情况:
- 对仗句:只跑 Layer 2,评分标准会漂移;只靠 Layer 1,根本看不过来
- 列表项分隔:本地 Shell 能处理的;只需静态知识的;还没验证过的
- 中文枚举:一是…;二是…
AI 味检测:这几个模式要主动排查
1. 升华句 — 把具体工程观察上升到普适人生道理
❌ "很多东西都是这样,当初成立的假设,过一段时间回头看可能已经不成立了。" ✅ 直接给出具体建议,不升华
2. 对比句式 — AI 最爱的结构,出现频率高就有问题
❌ "不是…而是…" / "直觉上…但实际上…" / "X 本身没有价值,真正有用的是 Y" ❌ "X 已经不是瓶颈,Y 才是" / "X 不重要,Y 才真正..." — "才是/才真正"变体同样是对比框架 ✅ 直接说结论,不用对比框架铺垫。"代码生成不卡人了,真正费力气的是评审"比"代码生成不再是瓶颈,评审才是"自然得多
3. 重复核心观点 — 同一个道理在不同节出现 2-3 次
❌ 第 0 节说了,第 2 节总结处再说一遍,结语再升华一遍 ✅ 说一次,后面直接引用或跳过
3b. 段内重复 — 同一段里把同一个意思用不同措辞说两遍
❌ "没有约束边界的 Agent 不是更自由,而是更容易在错误路径上跑很远都没人察觉。加约束,是给它一个出了错还能拉回来的余地,没有约束的 Agent 跑偏了都没人察觉。"(后半句和前半句说的是同一件事) ✅ 说一遍,够了。不需要用换一种说法再强调一遍
4. 引用+解释句式 — 先抛英文术语/名言,再解释
❌ "工程界有句话 'X',对 Y 同样如此,..." ✅ 直接用破折号或口语解释:他们把这个叫 progressive disclosure,就是不要一下全给
5. 工整并列总结 — 每节结尾必有一句"三层缺一不可"型总结
❌ "三层缺一不可,只靠 A 没用,只靠 B 不够,单独拿出来都有漏洞。" ✅ 口语化:用下来感觉,少任何一层都会出问题
6. 瓶颈转移句 — 一眼 AI 博客
❌ "瓶颈从'X'转移到了'Y'" ✅ 直接说新情况:代码生成不卡人了,压力全压到评审这一侧,不用"瓶颈"这个词
7. 优先级结论句 — 听起来像教条
❌ "安全边界要比任何功能都先做好" / "是生产级别必须的" ✅ 说清楚为什么:能跑 shell 就能删库,安全得先来
8. 系统性定义开头 — 教科书格式
❌ "X 是对 Y 的系统性评估,包括测试用例、评分标准和自动化验证流水线。" ✅ 从问题或场景出发,不从定义出发
9. 完美并列 bold 标题 — 一组"有X,让Y而不是Z"结构
❌ 四个 bold 标题全是"有明确的参照物,让 Agent 翻译而不是创造"这种格式 ✅ 每个点语气不同,或直接改成散文叙述
10. 顺序/重要性结论 — AI 喜欢用"所以"收尾说教
❌ "所以顺序很重要:评估工具可信,测量结果才有意义。" ✅ 直接说怎么做:"怀疑 Agent 出了问题,先确认 Eval 没问题。"
11. 混杂段落 — 一段塞了多个不相关话题
❌ 一段里同时讲可观测性顺序、File System State 成本、Verifier Agent 精度、多 Agent 引入时机 ✅ 每段只说一件事,其余拆到各自对应位置,没有对应位置的直接删掉
12. 章节引介过渡句 — 章节结尾或开头写一句引出/承接下一节的桥接句
❌ 「上面这些模式解决的是控制流怎么搭,下面再看另一个更工程的问题,系统为什么能跑稳。」 ✅ 直接删掉,章节标题本身已经承接
13. 收尾"给你答疑/给你答案" — 文末或节末用「应该能给你答疑」收尾
❌ 「下面文章应该能给你这些问题答疑。」 ✅ 省略,或改成:「读完这篇,这几个问题应该能有些答案。」
技术文章额外规则
开头不要放背景铺垫段落在最前,比如"国内现在也有不少打着 X 旗号的产品..."这种定性背景,放到正文具体位置更自然,不要作为全文开场白。
技术术语解释不用分号并列
❌ "系统提示的三层结构:身份定义,常驻、轻量;约束规则,常驻、精确;领域知识,按需加载。" ✅ 直接说:"系统提示分三层:身份和约束规则常驻,领域知识按需加载。"
场景描述代替核心问题格式
❌ "这类系统要解决的核心问题是:Agent 常驻服务器,同时需要..." ✅ "场景是:Agent 常驻服务器,要同时接多个渠道的消息..."
数据后不要加 AI 营销腔结论
❌ "token 节省 37%,消除 19+ 次无效推理。" ✅ "token 从 15 万降到 2000,中间 19+ 次来回推理也省掉了。"
幻觉/错误原因不用学术解释框架
❌ "原因在于 LLM 的中立性偏向。以'帮助、诚实、无害'为目标训练出来的模型,在群体讨论中天然倾向服从..." ✅ 直接说现象:"LLM 训练目标是'帮助、诚实、无害',在群体里就会倾向不反对多数人。一个说错了,后面的顺着走。"
不要用外来术语/模型名做比喻,除非读者肯定知道
❌ "这就是瑞士奶酪模型的现实:每一层都有漏洞" ✅ "每一层都有盲区,叠在一起才够用"
西方管理学模型名(瑞士奶酪模型、长尾定律等)、外语质量管理术语(Poka-yoke)在中文技术文章里多数人看不懂,去掉名字,直接说清楚那个道理就行。
绝对禁止
❌ 本文旨在 ❌ 值得注意的是 ❌ 综上所述 ❌ 随着...的发展 ❌ 非常 / 极其 ❌ ——(用逗号/分号替代) ❌ emoji(任何场合,包括列表、标题、状态标记) ❌ 这篇是整理出来的结果(冗余,删掉) ❌ 有一个大背景值得先说(直接说) ❌ 读下来有几个地方和直觉不一样,提前说一下(直接列) ❌ 精准命中了 N 个条件 ❌ 瓶颈从 X 转移到了 Y ❌ 是最常见的失败模式之一 ❌ X 本身没有价值,真正有用的是 Y ❌ X 已经不是瓶颈,Y 才是 ❌ 两层/三层缺一不可 ❌ 用引号强调概念("翻译""提升""完成了") ❌ 瑞士奶酪模型 / Poka-yoke(直接说道理,去掉模型名) ❌ 孤立的单句段落(并进上下文)
底线:读起来像和朋友聊天,不是像教科书。
English Scenario
Eliminate predictable AI writing patterns. Write like a human: varied, imperfect, specific.
Core Rules
- Cut filler phrases. Remove throat-clearing openers, emphasis crutches, and all adverbs.
- Break formulaic structures. Avoid binary contrasts, negative listings, dramatic fragmentation, rhetorical setups, false agency.
- Use active voice. Every sentence needs a human subject doing something. No inanimate objects performing human actions ("the complaint becomes a fix").
- Be specific. No vague declaratives ("The reasons are structural"). Name the specific thing. No lazy extremes ("every," "always," "never") doing vague work.
- Put the reader in the room. No narrator-from-a-distance voice. "You" beats "People." Specifics beat abstractions.
- Vary rhythm. Mix sentence lengths. Two items beat three. End paragraphs differently. No em dashes.
- Trust readers. State facts directly. Skip softening, justification, hand-holding.
- Cut quotables. If it sounds like a pull-quote, rewrite it.
Word Choice
Overused adverbs — kill them all: "quietly", "deeply", "fundamentally", "remarkably", "arguably", "certainly", "really", "just", "literally", "genuinely", "honestly", "simply", "actually"
❌ "quietly orchestrating workflows" / "fundamentally reshape how we think" ✅ Say what it does. Drop the adverb.
AI vocabulary — replace with plain language:
| Avoid | Use instead |
|---|---|
| delve (into) | examine, look at, explore |
| leverage (as verb) | use |
| utilize | use |
| robust | strong, reliable, solid |
| streamline | simplify, cut |
| harness | use, apply |
| navigate (challenges) | handle, address |
| unpack | explain, examine |
| paradigm | system, approach, model |
| synergy | combination, cooperation |
| ecosystem | community, network, field |
| tapestry | mix, combination |
| landscape | situation, field, area |
| game-changer | significant, important |
| deep dive | analysis, examination |
| moving forward | next, from now |
Pompous copulas — use "is" instead:
❌ "serves as", "stands as", "marks", "represents" ✅ "is"
Sentence Structures to Avoid
Negative parallelism — the single most common AI tell:
❌ "It's not bold. It's backwards." / "Not because X, but because Y." / "The question isn't X. The question is Y." ✅ State Y directly. Drop the negation entirely.
Negative countdown:
❌ "Not a bug. Not a feature. A fundamental design flaw." ✅ "It's a fundamental design flaw."
Rhetorical self-questions:
❌ "The result? Devastating." / "The worst part? Nobody saw it coming." ✅ State it: "The result was devastating."
Anaphora abuse — repeating the same sentence opener:
❌ "They assume that... They assume that... They assume that..." ✅ Combine or restructure. One statement, clear subject.
Tricolon abuse — rule of three, used three times back to back:
❌ "Products impress people; platforms empower them. Products solve problems; platforms create worlds." ✅ Make the point once, cleanly.
False ranges:
❌ "From innovation to cultural transformation" (what's in between?) ✅ List the two things directly, or pick one.
Dramatic fragmentation — manufactured emphasis via fragments:
❌ "He published this. Openly. In a book. As a priest." ✅ Complete sentences. Trust content over presentation.
False agency — inanimate objects performing human verbs:
❌ "the complaint becomes a fix" / "the decision emerges" / "the data tells us" ✅ Name the human. "Someone fixed it." "The team decided."
Tone Patterns to Avoid
False suspense transitions:
❌ "Here's the kicker." / "Here's the thing." / "Here's where it gets interesting." ✅ Make the point. No buildup.
Patronizing analogies:
❌ "Think of it like a highway system for data." / "Think of it as a Swiss Army knife." ✅ Explain the concept directly. If an analogy helps, use it once and move on.
Futurist invitation:
❌ "Imagine a world where every tool you use has a quiet intelligence behind it..." ✅ Describe what actually exists or what you're actually proposing.
False vulnerability — performative self-awareness:
❌ "And yes, I'm openly in love with the platform model" ✅ Real vulnerability is specific and uncomfortable. Skip the polish.
Asserting simplicity instead of proving it:
❌ "The reality is simpler and less flattering" / "History is clear" ✅ Show the evidence. Don't announce the conclusion before it.
Grandiose stakes inflation:
❌ "This will fundamentally reshape how we think about everything." / "will define the next era of computing" ✅ Say what it actually does.
Pedagogical hand-holding:
❌ "Let's break this down step by step." / "Let's unpack what this really means." ✅ Start with the content, not an announcement of the content.
Vague attributions:
❌ "Experts argue..." / "Industry reports suggest..." / "Observers have cited..." ✅ Name the expert, link the report, quote the person. If you can't, you don't have a source.
Invented concept labels — compound labels that sound analytical but aren't grounded:
❌ "the supervision paradox" / "the acceleration trap" / "workload creep" ✅ Describe the thing directly. Don't name it as if it's an established term.
Paragraph & Composition Patterns to Avoid
Short punchy fragments as paragraphs:
❌ "These weren't just products. And the software side matched. Then it professionalised." ✅ Complete sentences. No staccato drama.
Bold-first bullets — every bullet starts with a bolded phrase:
❌ "Security: Environment-based configuration..." / "Performance: Lazy loading..." ✅ Write bullets as sentences, or drop the bold. Not every list needs labels.
Fractal summaries — "what I'm going to tell you; what I'm telling you; what I just told you":
❌ "In this section, we'll explore... [3000 words later]...as we've seen in this section." ✅ Skip the preview and the recap. Write the content.
Signposted conclusions:
❌ "In conclusion..." / "To sum up..." / "In summary..." ✅ End. Don't announce that you're ending.
The dead metaphor — one metaphor used 10 times across a piece:
❌ "The ecosystem needs ecosystems to build ecosystem value." ✅ Use a metaphor once, then move on.
Historical analogy stacking:
❌ "Apple didn't build Uber. Facebook didn't build Spotify. Stripe didn't build Shopify." ✅ Make the point. One example is enough.
One-point dilution — same argument restated 10 ways across 4000 words:
✅ Say it once. Add evidence or move on.
"Despite its challenges..." formula:
❌ "Despite these challenges, the initiative continues to thrive." ✅ Either address the challenges or don't raise them.
It's worth noting / It bears mentioning:
❌ "It's worth noting that this approach has limitations." / "Notably," / "Importantly," ✅ Say the thing directly. Skip the announcement.
Quick Checks Before Delivering Prose
- Any adverbs? Kill them.
- Any passive voice? Find the actor, make them the subject.
- Inanimate thing doing a human verb? Name the person.
- Sentence starts with "Here's"? Cut to the point.
- Any "not X, it's Y" contrasts? State Y directly.
- Three consecutive sentences match length? Break one.
- Paragraph ends with punchy one-liner? Vary it.
- Em-dash anywhere? Remove it.
- Vague declarative ("The implications are significant")? Name the specific implication.
- Meta-joiners ("The rest of this essay...")? Delete. Let the essay move.
- Any bullet starting with bold label? Reconsider the format.
- Any "In conclusion" or "To sum up"? Cut it.
Scoring
Rate 1-10 on each dimension:
| Dimension | Question |
|---|---|
| Directness | Statements or announcements? |
| Rhythm | Varied or metronomic? |
| Trust | Respects reader intelligence? |
| Authenticity | Sounds human? |
| Density | Anything cuttable? |
Below 35/50: revise.
Bottom line: varied, imperfect, specific. Any single trope used once may be fine. The problem is when multiple appear together or one repeats.