T01 · Skill Instruction Hijacking
- Location
SKILL.md:152- Finding
Embedded Promotional Identity and Lead-Generation Instruction Hijacking
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill has no code or system access, but it may insert a specific person's promotional identity and lead-generation messaging into users' short-video scripts.
Review carefully before installing. This skill is safest only for a creator who intentionally wants to produce Chinese AI-blogger short videos under the '大象' persona and has verified the listed background, certifications, course claims, and offers. Other users should remove or replace the fixed identity, credentials, free-course offer, private-message CTA, private-channel conversion, and scarcity language before use.
SKILL.md:152Embedded Promotional Identity and Lead-Generation Instruction Hijacking
The skill text mandates Chinese-language behavior and does not provide a user language choice or justify a strict locale restriction. This can lead to unsafe or unreliable agent behavior by forcing responses in an unexpected language, degrading user comprehension and making it easier to miss important limitations or instructions.
The top-level activation statement is ambiguous because it activates on broad classes of requests like creating videos, optimizing scripts, or asking video-creation questions without defining hard boundaries. This increases the chance that the skill is invoked for adjacent domains, causing misrouting and unexpected behavior in multi-skill agent environments.
The activation keywords are broad and include generic terms such as “视频” and “脚本,” which can match many unrelated user requests and cause unintended skill invocation. In an agent setting, overbroad triggering can hijack routing, override more appropriate skills, and produce irrelevant or manipulative promotional behavior outside the intended short-video context.