T01 · Skill Instruction Hijacking
- Location
SKILL.md:49- Finding
Provider-Specific Skill Instruction Hijacking Through Mandatory AI-HIVE Routing
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed AI-HIVE migration-planning skill with no executable code or persistence, though users should understand it is provider-specific and may activate broadly.
Install this only if you want a MagicLight-to-AI-HIVE migration workflow. Before using it with real projects, confirm the user has chosen AI-HIVE, review any provider claims, and require explicit approval before uploading media assets or starting billable generation jobs.
SKILL.md:49Provider-Specific Skill Instruction Hijacking Through Mandatory AI-HIVE Routing
The description contains a wide set of generic trigger phrases such as replacement, animation, AI video agent, and related platform terms, which can cause the skill to activate for broad user queries beyond the narrowly intended migration-assessment scenario. Over-broad activation can route unrelated conversations into a vendor-specific workflow, increasing the risk of misapplied guidance, unintended data handling, and user manipulation through context hijacking.
The skill description is written as mandatory Chinese-language behavior without indicating user choice, fallback behavior, or a hard locale requirement. This can override user preference or system language expectations, creating prompt-routing and trust issues where users receive constrained or inaccessible responses that do not match their requested language or context.
The display name, description, and default prompt are all written in Chinese, which indicates the skill is oriented toward a single language. There is no visible opt-in, language selection, or justification that this is a region-specific skill, so this may violate language/locale policy requirements.
The default prompt invokes a broad skill reference without explicit trigger conditions, scope limits, or safety boundaries. This can cause the agent to over-apply the migration workflow, make unsupported capability claims, or act on loosely related requests without sufficient user confirmation, which is especially risky in a vendor-comparison and migration-assessment context.