Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed Markdown research helper that uses online sources for user-requested research, with no evidence of hidden code, exfiltration, persistence, or destructive behavior.
Install only if you are comfortable with your research topics being used to query external search, academic, social-media, and news sources. Avoid entering secrets, unpublished proprietary plans, or personal data, and review any suggested shell commands before running them.
The manifest tag '帮我研究' is a highly generic phrase that overlaps with common user language, so the skill may be invoked unintentionally. In context, this is more dangerous than a simple UX bug because the skill can send queries to multiple external sources and has access to exec, broadening the blast radius of mistaken activation.
The skill explicitly aggregates data from web search, academic platforms, social media, and news sources, but it does not warn users that their prompts or derived queries may be transmitted to third-party services. This omission is security-relevant because users may include sensitive research topics, proprietary information, or personal data, and the broad/ambiguous triggers increase the chance this happens without informed consent.
The free-version limits explicitly say report export to PDF/Word is not supported. Yet the capability description says the skill supports 创建/查询/导出操作 (create/query/export operations), which directly conflicts with the declared limitation and creates ambiguity about whether export exists in the free edition.
The skill repeatedly describes the free version as Markdown-instruction driven and explicitly says API integration is not supported in the free edition. However, callback_url implies outbound webhook/API behavior for async completion, which contradicts the stated free-version capability boundary rather than merely omitting implementation detail.
This line again states support for 创建/查询/导出操作, while later FAQ text says the free version only supports Markdown output and not PDF/Word/HTML export. That is an active contradiction in the skill's own documentation about what operations the free edition can perform.
The declared trigger condition ('when data analysis, report generation, statistics, visualization are needed') is overly generic and likely to activate in many unrelated contexts. Because the skill performs multi-source collection from online platforms, ambiguous routing can expose user prompts to third-party services without clear intent or informed consent.
The trigger phrase is broad enough to match ordinary conversation, increasing the chance of accidental invocation. In this skill, accidental activation is more dangerous because the skill is designed to access external sources and the manifest includes powerful tools like exec, which can lead to unnecessary data disclosure or unintended agent actions.
No suspicious patterns detected.
