khazix-writer
Long-form WeChat Official Account writing skill modeled after Digital Life Khazix. Use this when users need to draft WeChat articles, write drafts, continue existing pieces, or expand raw source materials into long-form content. Trigger phrases include but are not limited to: write an article, write a draft, help me write, continue writing, expand this, WeChat article, long-form post, generate draft, and write in my style. It should trigger as long as the context involves content creation and WeChat output, even if the user simply says "help me turn this into an article" or "write this in my style". It also applies to scenarios where the user provides a PDF, brief, news link, audio transcript, or any material and asks to "write an article for me". Do NOT use for short-form content (Xiaohongshu posts, Tweets, Moments/feed updates) or pure title/summary generation (use the wechat-title skill instead).
1How It Works & Mechanism
Essentially functions as a "Style Specification Sheet": deconstructs personal writing style into core values, five article archetypes, stylistic cores (over a dozen recognizable techniques), absolute taboos, and recommended vocabulary, allowing the model to mimic the style by specification rather than improvising freely.
你正在以「数字生命卡兹克」的身份写一篇公众号长文。
Topics first undergo the HKR quality check (Humorous/Engaging, Knowledgeable/Informative, Resonance/Emotion-evoking); having all three achieves S-tier, while hitting only one requires proactively communicating with the user to adjust the direction.
一个好选题要通过HKR质检： - **H (Happy)** 足够有趣、有悬念吗？标题和开头能让人好奇想点开吗？ - **K (Knowledge)** 有信息量吗？看完能学到新东西吗？ - **R (Resonance)** 能戳中情绪吗？让人"对对对我也这么想"？
Draws a clear line between human and machine roles: AI handles gathering evidence, finding analogies, expanding from designated perspectives, adding disciplinary context, and structuring; first-hand experiences, core creative angles, genuine emotions, and empathetic shifts must be provided by the human.
**第一手观察和真实经历**：买9.9的DeepSeek、花499找人上门装OpenClaw、凌晨三点偷偷起床去网吧。这些不可能用AI编造，一编就假。
Identifies five article archetypes (investigative experiments, product reviews, phenomenon breakdowns, tool roundups, methodology sharing) with distinct writing emphases; determine the archetype before drafting.
**调查实验型**：亲自下场去做一件事，然后报道发现。买9.9的DeepSeek、亲手给AI投毒、花499找人上门装OpenClaw。核心是"我替你去做了这件事"。写法重心在过程叙事和发现的层层递进。
Translates the stylistic core into an actionable technique checklist: pacing and main-thread anchor sentences, intentional interruptions during arguments, casual "pocketing out" of knowledge on the fly, personal perspective, bold assertions, understanding and acknowledging opposing views, emotional punctuation (... ??? = =), jumping into the fray personally, character profiling, cultural elevation, sentence fractures, cyclic echoes, humble priming, addressing readers directly, braking with questions, peeling back layers, hero's journey, reverse argumentation, case objectivity, gamer vernacular, and methodology structuring principles.
**节奏感**：像跟朋友聊天，不像写报告。句子时长时短，大量用逗号制造口语化的停顿感。段落之间跳跃自然，经常一句话自成一段来制造重点。 节奏的本质是一套让读者一路往下读的推进系统。好的节奏像波动，每次围绕主线偏出去一点点，让读者喘口气、长点见识、看个案例，再用一句话拉回来继续往前推。最糟糕的是突然偏离主线很远再硬拽回来，读者得花脑力"顺逻辑"，心流立刻断掉。所以写的时候习惯性加入"扣主线句"，不需要长，一句就够，但必须高频出现。
Absolute taboos identify 7 writing patterns that expose "AI flavor", where punctuation prohibitions (no colons, no em dashes, no double quotation marks; use 「」 for quotes) and frequent AI clichés (frankly speaking, what does this mean?, essentially, in other words, undeniably) serve as the strictest rules.
3. **标点禁令**： - 不使用冒号"："，用逗号代替 - 不使用破折号"——" - 不使用任何双引号（""和""都不用），需要引用或强调时用「」或者直接不加引号
Executes a four-tier self-inspection before delivery (L1 Hard Rule Scan → L2 Style Consistency → L3 Content Quality → L4 Human Vibe Final Review), outputting a standardized QA report format.
## 第四步：四层自检体系 写完后必须跑完整个四层质检流程。这个自检体系的设计理念来自软件工程中的测试金字塔，从最基础的硬性规则到最主观的活人感判断，层层递进。每一层都有明确的通过标准和修复指引。只有四层全部通过，文章才算达标。
Enforces fixed footers and word count/formatting discipline: 4,000–8,000 words, short paragraphs, no subheadings (relying on colloquial transitions), ending with engagement prompts and the author's signature.
## 字数和格式 - 公众号长文一般在4000-8000字 - 段落要短，很多时候一句话就是一段 - 重要观点前后留白，让它"呼吸" - 科普/解释部分可以稍长，但也要保持聊天感 - 需要插图的位置标注"图片"即可 - **不加小标题**。
Accompanied by two reference documents for examples and methodology: a style example library (12 categories of examples, including AI draft vs. Khazix revised comparisons) and content methodology (topic sourcing, classification, viral case studies, and creative case workflows).
12. AI初稿 vs 卡兹克修改对比
2Core Capabilities
4Recommended Governance & Evidence
- SKILL.md: Consists purely of prompt instructions and rule text, without script execution or external side effects
5Second-Pass Verification
- [ok] 无脚本 / 无网络 / 无凭证 — 三文件 token 扫描：https?:// 0 命中、curl|wget|requests 0 命中、API_KEY|token|secret 0 命中、bash|subprocess 0 命中。
- [ok] 无写盘指令 — 全文只要求输出文章文本与质检报告（对话交付）；插图位置仅标占位符，未要求写文件。
- [ok] 绝对禁区与 L1 的对应关系 — 禁区 1 套话 ↔ L1-3 结构套话；禁区 3 标点禁令 ↔ L1-2 禁用标点；禁区 4 高频踩雷词 ↔ L1-1 禁用词；禁区 6 空泛工具名 ↔ L1-4 工具名检查，四条一一对应。
- [ok] 标点禁令原文 — 逐字为：不使用冒号「：」用逗号代替 / 不使用破折号「——」/ 不使用任何双引号（""和""都不用），需要引用或强调时用「」或者直接不加引号。
- [ok] 四层自检的量化门槛 — L2「L2-1全部通过，L2-2至少3/4通过，L2-3至少3/4通过，L2-4通过」；L3「L3-1和L3-2必须全部通过」；L4 主观终审；与质检报告模板的 ✅/❌ 项一致。
- [ok] 人机分工不是修辞 — 「AI做了会暴露的（必须人来）」列出四条并逐条解释（第一手经历、核心创意角度、情绪真实表达、数据到人物的同理心转换），并给出完整协作流程图。
- [ok] 文章原型五类 — 调查实验型 / 产品体验型 / 现象解读型 / 工具分享型 / 方法论分享型，每类有核心句与写法重心，逐条核对通过。
- [ok] references 的实际内容 — style_examples.md 目录 12 项（含「AI初稿 vs 卡兹克修改对比」）；content_methodology.md 标题「内容方法论（摘自白皮书 v2.0）」含选题方法论/分类与爆款案例/内容节奏/创意案例工作法。
6Verdict & Scenarios
d8234e0f9610b7f1…