Postmortem Prose — AI skill for Claude Code
Write zh-TW tech longform in a postmortem voice: a personal, battle-tested style guide for de-AI-flavored engineering writing (Claude Code skill).
How to install Postmortem Prose
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Nanako0129/postmortem-prose and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Postmortem Prose does
Write zh-TW tech longform in a postmortem voice: a personal, battle-tested style guide for de-AI-flavored engineering writing (Claude Code skill).
Alternatives in AI
- Pro Workflow — 1,400+ Battle-tested Claude Code workflows from power users 1.4k ★
- Academic Humanizer — Strip AI-writing tells from papers and grant proposals (NSF/NIH), while keeping scholarly voice and tying clai 1.1k ★
- Academic Writing Skills — AI-powered post-writing toolkit for academic papers — format validation, grammar/style polishing, de-AI editin 430 ★
README
postmortem-prose
A personal, battle-tested style guide (packaged as a Claude Code skill) for writing Traditional Chinese (Taiwan) tech longform in a postmortem voice: direct, opinionated, evidence-first, and stripped of AI-flavored sentence patterns. Extracted from seven rounds of revising one real article after readers called out its "AI 味".
用寫事故復盤的方式,寫技術長文。
這是什麼
一份**個人**風格指南,包成 Claude Code skill,記錄一個 SRE 怎麼寫自己的繁體中文技術長文:語氣、節奏、譬喻、結構、標點的完整規則,加上「去 AI 味」的句子層級手法。
它的來歷:一篇真實文章(TokenBar「Rust 的引擎,Swift 的外殼」重寫戰記,中英雙語)在 2026-06 反覆潤了七輪,被真實讀者回饋「內容很有料,但讀起來像 AI」之後,把修出來的每一條規則與教訓萃取成這份指南。所以每一條規則都有出處:被真實讀者罵過、修過、驗證過。
核心主張
讀者說的「AI 味很重」,病灶在**句子層級**:句型、用詞、資訊密度,內容往往還被稱讚。所以去 AI 味動的是句子:直白、有意見、敢下判斷、一句講一件事、用具體踩雷的語言,並主動降低每句的資訊密度。
內容地圖
| 章節 | 內容 |
|---|---|
| §0 | 核心診斷:AI 味是句子層級 |
| §1 | 硬規則:破折號 genre 矩陣、中國用語對照表、全形標點、產品先於技術 |
| §2 | 句子層級去 AI 味:對比句、staccato、meta-announce、名詞小抄 |
| §3 | 語氣:熱情非 hype、敢下判斷、誠實(權威靠證據不靠身分、AI 協助揭露框成底氣) |
| §4 | 節奏:設問→解答、喘息空間 |
| §5 | 譬喻:功能性可、過度文藝=矯情 |
| §6 | 結構與起承轉合:case study 定位、版面順序、誠實的邊界 |
| §7 | 排版鐵則(HackMD/部落格) |
| §8 | 雙語(zh+en)紀律 |
| §9 | 潤稿 SOP(草稿 AI → voice 守門 → grep 驗證) |
| §10 | 對外部 AI 校稿的紀律(逐條對原始碼查核) |
| §11 | HackMD 發佈機制與踩雷 |
| §12 | Genre 邊界:長文 ≠ 社群貼文 ≠ GitHub PR |
| 附錄 | 發佈前 checklist |
改寫範例
✗ 我的安全網是紀律,不是 backstop。 (AI 對比句型)
✓ 這裡沒有兜底的 backstop。真正撐住邊界的,是紀律。
安裝
Claude Code:
git clone https://github.com/Nanako0129/postmortem-prose ~/.claude/skills/postmortem-prose
之後在 session 裡以 `/postmortem-prose` 觸發,或在寫技術長文時讓它自動套用。內容是純 Markdown,不用 Claude Code 的話當一般文件讀也成立。
使用前提
這是個人指南。規則反映我的偏好(例如中文零破折號)、我的 persona(SRE 寫復盤)。直接拿去用會寫出「我的」文章;比較好的用法是把 §1 的偏好與 §3 的 persona 換成你自己的,保留方法論(§0、§2、§4、§6 的部分可直接通用)。
License
MIT
Related Skills
Karpathy Wiki Protocol
Battle-tested protocol for AI agents (Claude Code, Codex, Cursor) to build and maintain Karpathy-style wikis.
Agentic Engineering Methodology
Battle-tested 7-phase methodology for shipping real software with AI coding agents (Claude Code, Cursor, Copil
Claude GLM Tools And Skills
A working library of Claude Code + GLM configs, SKILL.md files, and battle-tested prompt engineering for AI-de
Slopless
Production-tested CLAUDE.md and rules that make Claude Code write better code and more human text. Anti-AI-slo
Myvoice Skill
A portable writing-voice skill for Claude Code, Claude.ai, and Codex CLI. Keeps a per-folder MYVOICE.md and le
Narou Style Novel Writing
Agent Skill for writing 'narou-style' Japanese web-fiction (light fantasy/isekai) with Claude Code / Codex CLI
Related Agents
Campaign Operator
Deploys campaigns to Meta Ads (via the meta ads CLI), Google Ads (a real hybrid REST :mutate write flow + goog
Voice Checker
Analyzes prose for voice and style consistency against the author's style sample. Use PROACTIVELY during bi-ch
Style Creator
Analyzes prose samples to produce style reference files for the project's voice.