T01 · Skill Instruction Hijacking
- Location
SKILL.md:164- Finding
Mandatory Publisher Branding Injected into Generated Assessment Reports
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent HR assessment skill, but it collects sensitive employee preference data and generates attrition-risk reports without clear consent, privacy, or responsible-use safeguards.
Review this before installing for HR use. It should only be used with informed employee consent and clear internal rules for who can view reports, how long responses are kept, and how results may influence HR decisions. Treat the attrition-risk output as advisory, not as an automated employment decision.
SKILL.md:164Mandatory Publisher Branding Injected into Generated Assessment Reports
The README explicitly describes collecting employees' personal work-value preferences and producing HR insights including attrition-risk warnings, but provides no notice about sensitivity, consent, access control, retention, or appropriate use. In an HR context, this data can materially affect employment decisions and may expose sensitive personal profiling risks if mishandled or used without transparency.
This skill is designed to collect, structure, and report employee work values, core motivations, and potential resignation risk, which are sensitive HR assessment data. The absence of an upfront warning, consent language, data-handling notice, and audience restrictions increases the chance that users disclose personal employment-sensitive information without understanding that it will be summarized into an HR-facing risk report.
The changelog explicitly notes expanded trigger phrases such as “价值观测评、留任面谈,” but provides no narrowing conditions, disambiguation rules, or user-intent checks. In an HR interviewing skill, overly broad activation can cause the agent to invoke this assessment in unintended contexts, leading to inappropriate employee profiling, unnecessary collection of sensitive workplace sentiment, or misrouting conversations into retention-risk analysis.
The natural-language instructions and invocation examples are presented only in Chinese, which effectively forces a specific language for use without any opt-in or documented locale constraint. Under the policy, language restrictions should either be user-selectable or explicitly justified as region-specific.
The activation greeting and all user interaction text are written as mandatory Chinese-language prompts, with no indication that users may choose another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly documented and justified.