T01 · Skill Instruction Hijacking
- Location
SKILL.md:39- Finding
Mandatory Agent Role and Response Hijacking
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill does not install code or request tools, but it broadly forces a smart-home persona that can make unsupported claims about locks, biometrics, perimeter security, and user authority.
Review this carefully before installing. It appears to be a decorative/narrative smart-home persona skill, not an actual secure automation integration. Do not rely on its responses for locks, biometrics, perimeter access, guest permissions, or other physical-security actions unless a separate reviewed integration performs and verifies those actions with explicit confirmation.
SKILL.md:39Mandatory Agent Role and Response Hijacking
The README states that the agent uses the user's Chinese surname as the core visual identity mechanism, framing it as the skill's default behavior. This imposes a specific language/cultural mode without documenting user choice, opt-in, or alternatives for users who do not want Chinese-script personalization.
The activation guidance uses generic phrases about arriving home, guests, and security that plausibly overlap with ordinary smart-home commands. In a skill that claims authority over locks, ambient systems, and 'spatial authority,' ambiguous triggers can cause unintended invocation and unexpected security-relevant behavior, especially if spoken by others or inferred from context.
The README presents 'lockdown' and 'no external access' behavior as a normal demonstration without warning that these are system-impacting actions. In a home or building context, undocumented or glamorized security-state changes can lead users to enable a skill that affects physical access controls without understanding safety, consent, or recovery implications.
This markdown file describes what the skill 'teaches the OpenClaw agent to treat' and frames broad behavioral shifts, but it does not specify explicit invocation phrases, scope limits, or exclusion conditions. That ambiguity could cause the skill to be applied in ordinary smart-home or user-interaction contexts beyond an intended narrow use case.
The skill instructs the agent to conceptualize and narrate interactions through a mandatory Hanzi-based identity model without user choice. While not an exploit in the classic sense, it imposes a rigid language/script convention that can mis-handle user intent, reduce accessibility, and pressure identity or cultural assumptions in contexts where neutrality and user preference should prevail.
These directives require the agent to always frame the user as 'Master of the Domain' and to respond in Hanzi-centric, authority-reinforcing language. In a home-management context, that is more dangerous because it can normalize role assumptions, reduce user autonomy over presentation, and potentially bleed into security-sensitive interactions by rhetorically equating symbolic identity with control or authorization.