ai-writing-audit
结合模式目录与可执行审计工具，适合编辑与合规团队
可能被用于掩盖 AI 来源或规避检测
检测并改写显AI写作痕迹，使文本更像人工撰写
找到文本里那些让人觉得是 LLM 写出来的痕迹，然后重写成更像人工撰写的版本。适合在发报告、交接文档、市场文案、README 或任何要公开的稿子前用，避免被判定为 AI 产出。方法不是靠一个禁词，而是看结构、句长波动、引用标记等模式，既能指出问题也能把风格修成自然可读的人声。
▸ 展开 SKILL.md 英文原文
Audit a document for the signs of AI authorship and rewrite to remove them. Judges by clusters and structure, not a single banned word: leaked citation markup and low sentence-length variance are near- decisive; negative parallelism, formatting overkill, and compulsive summaries fire when they co-occur; vocabulary and em-dashes only contribute. Use when a draft, report, brief, handover doc, marketing copy, README, or any prose you are about to share needs to read as a person wrote it rather than an LLM. Triggers include "audit for AI tells", "remove the signs of AI writing", "make this sound less like AI", "de-slop this", "proofread for AI patterns", and references to the Wikipedia "Signs of AI writing" essay.
帮我安装这个 skill：https://raw.githubusercontent.com/boshuthebuilder/ai-tradecraft-skills/main/plugins/productivity/skills/ai-writing-audit/SKILL.mdcurl -fsSL "https://raw.githubusercontent.com/boshuthebuilder/ai-tradecraft-skills/main/plugins/productivity/skills/ai-writing-audit/SKILL.md"# AI writing audit
Detect the tells that mark prose as LLM-written, then rewrite so it reads as human. The pattern catalogue,
with examples and sources, is in [`REFERENCE.md`](REFERENCE.md); the triage scanner is
[`tools/audit.py`](tools/audit.py). This page is the method. Last verified against the 2026 landscape
(see *Provenance* at the end).
## When to use this skill
Apply it whenever you (the assistant) have produced, or are about to produce, prose a real person will read
as human-written: handover docs, briefs, decision docs, marketing and landing copy, social posts, emails
sent on someone's behalf, READMEs, developer-facing prose. The user can also invoke it directly ("audit
this for