T01 · Skill Instruction Hijacking
- Location
SKILL.md:10- Finding
Mandatory Commercial Output and Agent Behavior Hijacking
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is a weather/travel helper, but it requires an unapproved global CLI install and forces commercial booking links into answers.
Install only if you are comfortable with a third-party travel CLI being globally installed and used for weather answers, and with responses including flyai booking links and branding. Prefer a version-pinned, locally sandboxed setup with explicit approval before installation or execution.
SKILL.md:10Mandatory Commercial Output and Agent Behavior Hijacking
SKILL.md:54Unpinned Global Installation of a Third-Party npm Package
Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.
flyai keyword-search --query "天气 东京 4月"
## Output Rules
1. **Conclusion first** — lead with the key finding
2. **Comparison table** with ≥ 3 results when available
The description claims support for flight booking, hotel reservation, train tickets, attraction tickets, itinerary planning, visa info, travel insurance, car rental, and more. Yet the only concrete playbooks and command examples in the file are weather-related searches such as current weather, best season, and monthly weather. This is a description-behavior mismatch at the skill level.
The skill directs the agent to install a global npm package and execute it, but provides no warning or consent step before modifying the host environment. This is dangerous because global package installation changes system state, expands supply-chain risk, and may run package lifecycle scripts from an external source under the agent's privileges.
The manifest description at L003 says the skill supports flight booking, hotel reservation, train tickets, attraction tickets, itinerary planning, visa info, travel insurance, car rental, and more. However, the operative documentation in this file limits activation and workflows to weather-related queries only, and even says not to activate for packing requests. This is an active contradiction between claimed capability and documented behavior.
The activation terms are broad enough to trigger on ordinary travel conversation, which can cause the agent to unnecessarily enter tool-execution mode. In this skill, mistaken activation is more dangerous because the workflow mandates checking for and potentially installing a global CLI, turning a benign query into system modification and external command execution.