Security audit
Security checks for vulnerabilities and agentic risk
This is a straightforward X/Twitter search helper that sends searches to Redfox; I found no hidden code, persistence, or unrelated authority.
Install this only if you are comfortable using a Redfox API key and sending your X/Twitter search terms and pagination requests to redfox.hk. Use explicit X/Twitter search wording, and be aware that short follow-ups like “continue” after a search may load another page.
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
---
name: twitter-work-search
description: "输入关键词搜索X(Twitter)作品，支持按热门、最新、媒体、用户、列表五种类型精准筛选，一键获取作品正文、互动数据、媒体附件与作者信息等结构化结果。触发词：当用户提到'X搜索'、'Twitter搜索'、'作品搜索'、'X关键词搜索'、'搜一下X上的'时使用"
---
The phrase 'Enter any keyword to instantly search tweets, users, and lists on X' describes an unrestricted trigger scope without clarifying limits, supported query forms, or non-matching cases. In documentation for a markdown skill file, this can make invocation criteria too open-ended and prone to unintended activation.
The README says users can 'Just describe what you need in natural language' with 'no commands to memorize,' but it does not define clear trigger boundaries or exclusion conditions. This broad phrasing could overlap with ordinary conversational requests and make it unclear when the skill should activate versus when it should not.
The README explicitly says users can interact using unconstrained natural language, which broadens the activation surface and makes accidental or unintended triggering more likely. In an agent setting, vague activation guidance can cause the skill to activate on loosely related user phrasing, increasing the chance of inappropriate tool use or data retrieval without clear user intent.
The paging triggers include extremely common phrases like '继续', '还有吗', and '下一页', which are likely to appear in ordinary conversation outside a deliberate paging request. This creates a realistic risk of unintended pagination and continued data retrieval, especially in multi-turn chats where such phrases may refer to something else.
The skill description embeds trigger phrases that are broad enough to match common user language, such as general requests to search X/Twitter content. This can cause unintended activation in unrelated conversations, leading the agent to invoke this skill without clear user intent and potentially exposing external-search results or consuming API-backed actions unexpectedly.
Using generic pagination cues like '继续' and '还有吗' creates a context-confusion risk because these phrases are common in ordinary dialogue and may trigger follow-on API calls without the user specifically asking to paginate search results. In a stateful assistant, this can lead to unintended data retrieval, extra cost, and surprising behavior across turns.
The workflow instructs the agent to send user-provided search keywords and an account-linked API key to a third-party endpoint without any explicit privacy notice or consent step. Even if the transmission is functionally required, users may not realize their queries, pagination state, and potentially sensitive interests are being disclosed to an external service, creating privacy and compliance risk.
No suspicious patterns detected.