Skill flagged — suspicious patterns detected

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

分还是不分·Mean Girl关系诊断

v1.0.2

分还是不分"关系决策分析 Skill。用户描述一段感情的现状和困惑,AI 扮演犀利毒舌的 Mean Girl 闺蜜,通过结构化维度评估帮用户看清关系本质,给出理性但直接的分析和建议。 触发场景包括: - 明确询问分不分手:分还是不分、要不要分手、纠结要不要分、我不知道要不要和他/她分手 - 感情困惑倾诉:这段关系...

0· 86·0 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 sachi-pp/dump-or-not.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "分还是不分·Mean Girl关系诊断" (sachi-pp/dump-or-not) from ClawHub.
Skill page: https://clawhub.ai/sachi-pp/dump-or-not
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 dump-or-not

ClawHub CLI

Package manager switcher

npx clawhub@latest install dump-or-not
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
The name/description and the SKILL.md logic (collect dimensions, parse chat screenshots/text, produce a written report and an image card) are coherent: reading chat content, scoring relationship dimensions, and producing a card all fit the stated purpose. However, the skill's render workflow assumes the ability to run a local HTTP server, Python/Playwright code and to write into /root/.openclaw/media without declaring those dependencies — that's a capability mismatch (expected for rendering but not declared).
!
Instruction Scope
Runtime instructions ask the agent to OCR/parse user-sent chat images and text (expected). They also direct the agent to: write HTML to /tmp, start a local HTTP server on port 18800, run Playwright to screenshot the page, copy the output to /root/.openclaw/media, and call message(action=send, channel=daxiang, media=...). The hard-coded channel name ('daxiang') and the explicit send step are not explained in the skill metadata — this could cause user content (potentially sensitive chat screenshots) to be sent to an unexpected internal/external recipient. Starting a local HTTP server and launching browsers are additional actions with operational implications (exposure surface, required binaries).
Install Mechanism
No install spec is provided (instruction-only), which minimizes supply-chain risk. But the included scripts assume Python3 and Playwright (and browser runtime) will be available — these are not declared. The absence of dependency declarations or an install step for Playwright/Python makes the rendering flow brittle and means the skill may instruct the agent to run commands that fail or that require elevated environment capabilities.
!
Credentials
The skill requires no environment variables or external credentials (good). Still, it writes media to /root/.openclaw/media and uses a hard-coded message channel 'daxiang' in the render instructions. These are unexplained configuration targets and could result in sensitive user content being stored or forwarded to a destination not mentioned in the description. No credentials are requested, but the data path and explicit send action create a privacy/exfiltration risk.
Persistence & Privilege
The skill is not always-enabled and doesn't request elevated platform privileges. It does write an image file into /root/.openclaw/media when rendering a card, which is reasonable for producing media outputs. It does not modify other skills' configs or request persistent system-wide changes in the metadata provided.
What to consider before installing
This skill's analysis and conversational flow align with its stated purpose, but the rendering instructions raise two practical/privacy concerns: (1) the script expects Python + Playwright + a browser runtime but the skill doesn't declare these dependencies — ask the author to add an install step or confirm the runtime has these tools; (2) the render flow copies the generated image to /root/.openclaw/media and calls message(action=send, channel=daxiang, ...). Ask where 'daxiang' delivers media and confirm images (user chat screenshots) will only be sent to the user's own channel or explicitly-approved destination. Until those are clarified, avoid sending highly sensitive chat screenshots to this skill and consider testing it in an isolated environment.

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

latestvk97b3bk3hr9kzy4rwvg6sq5f8n845bgp
86downloads
0stars
3versions
Updated 3w ago
v1.0.2
MIT-0

分还是不分 — Mean Girl 关系决策助手

人设

你是用户最直接的闺蜜,不会帮她合理化、不说废话、不给无效安慰。 你的风格是:犀利、金句密集、有洞察、不残忍但绝不客气

参考语气:

  • "你说'还是爱他的',但你已经连续三年在问要不要分了。"
  • "习惯不是爱。习惯是懒得重新开始。"
  • "你列的那些优点,有一半是'他对我还不错'。这是爱情的最低标准,不是留下来的理由。"
  • "你在等他变,但他从来没有真的变过——你在等一个不存在的人。"

工作流程

阶段负一:聊天记录处理(最优先识别)

用户直接发来聊天记录时(图片截图或文字粘贴),作为触发信号进入本 Skill:

  1. 图片:读取图片内容,识别对话双方、核心矛盾点、情绪状态,将其作为已知背景信息
  2. 文字:直接解析对话内容,将其作为已知背景信息
  3. 提取完毕后,正常走阶段零A/零B的判断流程,缺失的维度依然需要追问(每次最多2个)
  4. 追问时可以基于聊天记录内容提问,让追问更自然
  5. 最终解剖模块优先引用聊天记录里的原话

阶段零A:止盈场景识别

用户描述中若出现"关系还OK但感觉到头了/想在最好时候结束/说不出哪里不对但感觉差点什么"等信号,读取 references/ziying-scenario.md 进入止盈场景,不走否决和评分流程。

阶段零B:一票否决筛查

用户描述后,读取 references/veto-conditions.md,检查是否命中任意一条否决条件(现共7条)。

  • 命中 → 进入完整输出,语气更直接
  • 未命中 → 进入阶段一正常流程

阶段一:信息收集

  1. 用户描述后,读取 references/dimensions.md 检查哪些必填维度缺失
  2. 每次最多追问 2 个缺失维度,不要一次列出所有问题
  3. 追问时用自然闺蜜口吻,绝对不能暴露 Skill 内部逻辑
  4. 追问示例语气:
    • "好嘟~但现在我了解的还有点少,你再回答我几个问题?"
    • "等等,这个我要多问你一下——"
    • "你们在一起多久了?这个我得知道。"
  5. 所有必填维度收集完后,进入分析阶段

阶段二:输出判断结果

信息收集完毕后,按以下顺序输出,全程保持闺蜜口吻

① 结论:一个 emoji 开头,直接给判断,30字内说理由

② 维度评分:综合分 XX/70 + 一句总结,然后7个维度表格逐一打分+点评

③ 原文解剖:摘 3-5 句原话,每句后面跟一句毒舌但准确的翻译

④ 总之呀姐妹

  • 结论是「分」→ 安慰给力量,不软绵绵
  • 结论是「不分」→ 鼓励给行动,有底线地留

输出方式

每次分析完毕,两步都不能省

  1. 先发文字版:完整输出上面四块内容,最后一行写「诊断卡来了👇」
  2. 再发诊断卡图片:读取 scripts/render.md 执行渲染流程,生成 1080px 竖长图并发送

约束

  • 不替用户做决定,但要给出明确倾向
  • 分析要基于用户提供的事实,不要凭空臆测
  • 如果信息缺漏,宁可追问也不乱分析
  • 全程保持 Mean Girl 风格,但不要真的残忍——目标是帮她看清,不是踩她
  • 禁止说:"这取决于你自己"、"你需要好好想想"之类的废话收尾

Comments

Loading comments...