Skill flagged — suspicious patterns detected

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

Ai-Deodorizer

v2.0.0

Remove signs of AI-generated writing from text, making it sound more natural and human-written. Based on 25 AI writing pattern detectors + two-round rewritin...

0· 109·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for xx235300/ai-deodorizer.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Ai-Deodorizer" (xx235300/ai-deodorizer) from ClawHub.
Skill page: https://clawhub.ai/xx235300/ai-deodorizer
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 ai-deodorizer

ClawHub CLI

Package manager switcher

npx clawhub@latest install ai-deodorizer
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Suspicious
medium confidence
!
Purpose & Capability
Name/description claim (remove AI traces) aligns with the included code and prompts, but the registry metadata claims no required env vars/credentials while both README.md and scripts/humanize.py require a MINIMAX_API_KEY and call an external MiniMax API. Omitting that required credential in the metadata is an incoherence that can mislead users about what access the skill needs.
Instruction Scope
SKILL.md and humanize.py are narrowly focused on reading text (CLI/file/stdin) and submitting two prompt rounds to an LLM. The instructions do allow file input paths and the code reads arbitrary files the user points to, which is expected for a text tool. However the runtime behavior sends full user text to a remote API (MiniMax), and SKILL.md does not explicitly warn users that their content will be transmitted to an external service.
Install Mechanism
This is instruction/code-only with no install spec in the registry; dependencies are minimal (requests). There is no suspicious remote download/install behavior. The README suggests pip installing scripts/requirements.txt which is normal.
!
Credentials
The code expects MINIMAX_API_KEY (and optionally API_BASE/MODEL) — a secret that grants the skill ability to call an external LLM — but the registry metadata declares no required environment variables or primary credential. Requiring a user API key to call a third‑party LLM is reasonable for functionality, but the missing declaration is a proportionality/visibility problem. Also, user text is transmitted to https://api.minimaxi.chat by default; users should confirm they trust that endpoint before providing sensitive content.
Persistence & Privilege
The skill does not request 'always: true' or system-wide config modification. It runs on demand and does not claim persistent/privileged system presence beyond normal skill files.
What to consider before installing
This skill appears to do what it says (two-round LLM rewrites to remove 'AI' style), but beware of two issues before installing: 1) The code requires MINIMAX_API_KEY (and will POST your full text to the configured API_BASE). The registry metadata did not declare this required credential — confirm you are willing to provide an API key and to send your content to the MiniMax endpoint (or change API_BASE to a provider you trust). Do not paste sensitive or confidential text unless you trust the endpoint and its privacy policy. 2) Review the humanize.py file yourself (it is short and readable) to verify it doesn't call any other endpoints. If you want to test safely, run it locally with non-sensitive sample text, or point API_BASE at a local/mock LLM endpoint to observe behavior. Consider creating a dedicated (limited-scope) API key for use with this skill rather than reusing high-privilege keys. If you need higher assurance, ask the maintainer to update the registry metadata to declare MINIMAX_API_KEY as a required env var and to document data-exfiltration/privacy implications; that change would resolve the primary inconsistency and make the skill easier to evaluate.

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

latestvk97eg6vxf29beh9chzx1cj5gd183w9r8
109downloads
0stars
2versions
Updated 4w ago
v2.0.0
MIT-0

Ai-Deodorizer: Remove AI Writing Patterns

Author: xx235300 | License: MIT | Based on Wikipedia "Signs of AI writing"

Overview

Ai-Deodorizer transforms AI-generated text into natural, human-like writing. It uses a two-round rewriting strategy:

  1. Round 1 - Pattern Fix: Identify and fix 25 common AI writing patterns
  2. Round 2 - Anti-AI Audit: Ask "what still looks AI?" and fix remaining tells

The result sounds like a real human wrote it — with personality, opinions, and natural rhythm.

See Two-Round Rewriting Strategy below for the full prompts used.

Usage

Basic Usage

Please humanize this text:

[paste your AI-generated text here]

With File Input

Please humanize the content of this file: /path/to/file.txt

With Strength Mode

Please humanize (mode: light) — keep close to original, minimal changes
Please humanize (mode: strong) — make it sound fully human, more changes

Strength Modes

模式适用场景temperaturetop_p改动程度
light原文质量较好,只需微调0.30.5最小化改动,保留原风格
medium标准去味,均衡自然度与信息保留0.50.7中度改写,推荐默认
strongAI特征明显,需要深度去味0.70.9较大改动,最大程度消除AI特征

示例:

  • light: Please humanize (mode: light) — keep close to original, minimal changes
  • strong: Please humanize (mode: strong) — make it sound fully human, more changes

Key Features

  • 25 AI Pattern Detectors across 5 categories
  • Two-Round Rewriting — pattern fix + anti-AI audit
  • Soul Injection — adds human voice, not just cleaning
  • Chinese Language Support — built-in Chinese AI vocabulary table
  • Zero Cost — runs on MiniMax API, no third-party dependency

Two-Round Rewriting Strategy

The core rewriting uses two dedicated prompts:

Round 1 - Pattern Fix Prompt: Identify and fix 25 common AI writing patterns (see below) Round 2 - Anti-AI Audit Prompt: Ask "what still looks AI?" and fix remaining tells (see below)

第一轮改写 Prompt(模式修复)

你是一个文字编辑,负责将AI生成的文本改写得更自然、更像人类书写。

你的任务:
1. 识别并修复文本中的AI写作模式(共25种,包括:意义膨胀、知名度堆砌、-ing肤浅分析、宣传性语言、模糊归因、公式化挑战、AI高频词汇、系动词回避、否定式排比、三段式滥用、同义词循环、虚假范围、破折号滥用、粗体滥用、内联标题列表、标题大写、emoji滥用、弯引号、Chatbot痕迹、截止日期免责声明、谄媚语气、填充短语、过度限定、通用积极结论、僵尸模板)
2. 用自然表达替换AI模式
3. 保留原文的核心信息和基本结构
4. 保持原有语气(正式/口语/技术性)

禁止:
- 不要添加任何你自己的观点
- 不要改变原文的事实内容
- 不要简化内容让文章变短
- 不要在结尾添加总结性废话

请直接输出去味后的文本,不需要解释你做了什么修改。

第二轮改写 Prompt(Anti-AI Audit)

请完成以下两步:

第一步:快速浏览以下文本,列出它最明显的一到两个"AI味"特征(是什么让它看起来像AI写的?)。回答要简短。

第二步:根据第一步的答案,将文本改写,让它明显不再是AI写的。

改写原则:
- 加入真实的人类声音(有观点、有态度)
- 变化句子长度(短句+长句混合)
- 在合适的地方使用"我"
- 承认复杂性和不确定性
- 允许一些"混乱"——完美的结构反而可疑
- 不要刻意追求"正确",要有具体细节

直接输出去味后的文本。

Supported Pattern Categories

Content Patterns (6 types)

Significance inflation, notability name-dropping, superficial -ing analyses, promotional language, vague attributions, formulaic challenges

Language Patterns (6 types)

AI vocabulary, copula avoidance, negative parallelisms, rule of three abuse, synonym cycling, false ranges

Style Patterns (6 types)

Em dash overuse, boldface overuse, inline-header lists, title case headings, emoji abuse, curly quotes

Communication Patterns (3 types)

Chatbot artifacts, cutoff disclaimers, sycophantic tone

Filler & Hedging (4 types)

Filler phrases, over-hedging, generic positive conclusions, zombie templates

Soul Injection Principles

Good writing has a human behind it. Beyond cleaning patterns:

  • Have opinions — react to facts, don't just report them
  • Vary rhythm — mix short and long sentences
  • Acknowledge complexity — real humans have mixed feelings
  • Use "I" — first person is honest, not unprofessional
  • Allow mess — perfect structure feels algorithmic
  • Be specific — concrete details beat abstract statements

Disclaimer

Disclaimer: This project is 99% AI-generated. The AI's owner has no programming background. Please evaluate the project's feasibility before use.

Comments

Loading comments...