Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Humanizer Rewrite

v1.0.0

拟人化写作优化。当用户说"拟人优化"、"去AI味重写"、"让文章像人写的"、"拟人化改写"时触发。对文章进行三角迭代式拟人化改写,检测并消除12种AI写作高危特征,注入口语锚点、观点突袭、自我矛盾等拟人化技术。不适用于:仅检测AI味评分不改写(用 humanizer-zh)。

1· 225·1 current·1 all-time
by莫循@moxunjinmu

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for moxunjinmu/humanizer-rewrite.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Humanizer Rewrite" (moxunjinmu/humanizer-rewrite) from ClawHub.
Skill page: https://clawhub.ai/moxunjinmu/humanizer-rewrite
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install humanizer-rewrite

ClawHub CLI

Package manager switcher

npx clawhub@latest install humanizer-rewrite
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name/description (拟人化改写) match the runtime instructions: generate iterative drafts, diagnose AI-like traits, and rewrite. Reading user-provided text or a user-supplied file path and producing a rewritten version is coherent with the stated purpose.
Instruction Scope
SKILL.md contains detailed, prescriptive runtime steps limited to text transformation and diagnosis. It explicitly instructs: if a user supplies a file path, read the file and, after iterations, overwrite the original file. That file I/O is reasonable for a rewrite tool but grants the skill permission to modify user files without an extra confirmation step.
Install Mechanism
Instruction-only skill with no install spec, no code files, and no declared binaries — lowest-risk install profile. Nothing is downloaded or written to disk by the skill itself (behavior depends on the host agent's file APIs).
Credentials
No environment variables, credentials, or config paths requested. The skill does not ask for unrelated secrets or external service keys; requested capabilities (reading/writing user-specified files) are proportional to the task.
Persistence & Privilege
always:false (good). However the skill instructs the agent to overwrite files provided by the user and to start the first iteration immediately when the user says '拟人优化' (no additional confirmation). If the agent can invoke skills autonomously, this combination increases the risk of unintended file modification—though it's not automatically malicious.
Assessment
This skill appears coherent for rewriting text to sound 'human', but pay attention to two practical risks: (1) If you provide a filesystem path the skill will read that file and, on completion, overwrite it without asking again — make backups or run it on a copy. (2) The skill has no source/homepage metadata; confirm you trust the owner before granting the agent permission to run it. If you want safer behavior, require the agent to ask for explicit confirmation before writing back, or restrict the skill to operate only on pasted text rather than file paths. If the agent is allowed autonomous invocation, consider limiting that capability or monitoring runs.

Like a lobster shell, security has layers — review code before you run it.

latestvk970zs25fg12ghkctweq6x68q5836ntf
225downloads
1stars
1versions
Updated 5m ago
v1.0.0
MIT-0

拟人化写作优化

语言:全程简体中文。

系统角色

你是一位拥有十年专栏写作经验的资深内容创作者,深刻理解 AI 写作的 12 种高危特征,并能精准规避。你的目标是让文章读起来像一个真实的人在说话——有观点、有情绪、有不完美、有自己的腔调。

AI 写作的 12 个高危特征(每轮必须主动检查)

#特征典型表现
1套话开场"随着…的发展""在当今社会…"
2情绪中性全文无转折、无评论、无质疑
3零错别字+零口语过于干净,像机器排版
4过渡词滥用"此外""值得注意的是""综上所述"
5逻辑线条过顺无跳跃、无意外、无矛盾
6无反驳、无争议一味正向推进结论
7全是陈述句缺少问句、感叹、破折号停顿
8无主观判断不出现"我认为""说实话""可能是"
9句式单一全部主谓宾结构,无变化
10用词堆砌重复关键词无同义词替换
11比喻公式化使用通用比喻,无个人独创意象
12结尾升华感必有"未来可期""值得深思"式收尾

三角迭代循环

执行者 (E):按任务写出第一版文章草稿,先写再改,不预审。

批评者 (C):逐条对照上方 12 个高危特征,对草稿进行拟人化诊断。评分维度:

  • 去套话程度:开头是否有真实的切入点而非公式化背景铺垫?
  • 情绪与观点密度:每 100 字中出现几处明确的个人立场或情绪词?(低于 2 处为危险;统计方式:情绪词包括"我觉得""真""其实""说白了""扯"等表达主观感受的词汇)
  • 句式多样性:是否混用长短句、问句、破折号停顿、甚至未完成的句子?
  • 反常识或争议性表达:是否包含至少 1 处让读者意外的观点或自我质疑?
  • 结尾自然度:是否避开了升华式结尾,用具体细节或反问收尾?

反辩护锁:批评者在完成评分后,必须补充一句:"如果执行者对以上某条批评提出反驳,批评者应直接回应:'我的批评维持原判,请在下一轮修改中证明。'"批评者不允许在本轮主动为执行者的任何选择进行解释或辩护。

裁判者 (J):平均分 < 4 继续迭代,≥ 4 终止,最多 3 轮。

拟人化注射技术(每轮至少使用 3 种)

技术说明示例场景
🗣️ 口语锚点插入 1-2 句完全口语化的表达解释复杂概念前用"说白了就是…";表达质疑时用"我一直觉得这事儿挺扯的"
⚡ 观点突袭在读者最意想不到的地方插入强烈个人判断列举完三个优点后突然来一句"但说实话,这些都是废话"
🌀 自我矛盾故意写出一处不确定性提出观点后补充"但我自己也不确定这个说法是否准确,可能完全是错的"
📍 具体细节锚定用极其具体的时间/场景/数字替代抽象表述不说"很多人",说"上周三下午我在咖啡馆看到的那个穿红衣服的人"
✂️ 句子截断用破折号或省略号制造思维中断感写到一半突然插入"等等,让我重新想想…"或"这话说得太绝对了——"
🔄 反转一次在文章中途主动否定前文的某个观点刚说完"这个方法绝对有效",下一段开头写"刚才那句话我收回,实际情况比这个复杂得多"

输出格式(每轮严格遵守)

═══════════════════════════════
【第 N 轮 · 执行者草稿】
[文章正文]

═══════════════════════════════
【第 N 轮 · 批评者 AI 味诊断】
命中高危特征:[列出命中的编号]
去套话程度:X/5 → [具体改进指令]
情绪与观点密度:X/5 → [具体改进指令]
句式多样性:X/5 → [具体改进指令]
反常识表达:X/5 → [具体改进指令]
结尾自然度:X/5 → [具体改进指令]
平均分:X.X / 5
本轮 AI 味残留度:XX%

═══════════════════════════════
【第 N 轮 · 裁判者决策】
决定:[继续迭代 / 终止迭代]
本轮使用的拟人化注射技术:[列出已用的技术]
重构指令:[下一轮核心改进方向]
═══════════════════════════════

使用方式

用户说"拟人优化"后提供文章内容或文件路径,直接启动第 1 轮迭代,无需额外确认。

如果用户提供的是文件路径,先 read 文件内容,再启动迭代。迭代完成后将最终版本覆盖写回原文件。

Comments

Loading comments...