Security audit
Security checks for vulnerabilities and agentic risk
This is a Taiwan travel-planning skill with local preference memory and no code execution, network access, or hidden high-impact behavior found.
Before installing, be comfortable with the skill keeping Taiwan trip preferences such as dates, pace, transport style, food restrictions, and travel history in local memory if you opt in. For live conditions such as weather, closures, prices, schedules, or emergencies, ask the agent to verify current sources.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Depending on one app for every rail layer
- Waiting until station time to install or register
- Using bikes casually without checking local rules or setup requirements
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
- Depending on one app for every rail layer
- Waiting until station time to install or register
- Using bikes casually without checking local rules or setup requirements
The activation rule 'Read this when ~/taiwan/ doesn't exist or is empty. Start naturally.' is broad and underspecified, which can cause the skill to trigger in unintended contexts or without an explicit user request. While not directly harmful on its own, ambiguous activation increases the chance of unsolicited behavioral changes and makes it harder to reason about when the skill should govern the assistant's actions.
The skill instructs the agent to ask whether to keep Taiwan preferences for future trip ideas and then 'Save their answer to main memory,' but it does not clearly warn the user that this data may persist beyond the current session. This can lead to users disclosing travel preferences, constraints, or related personal data without informed consent about retention, creating a privacy risk through unexpected persistence.
The section titled "How to Order or Talk About Tea" presents a single "Best language" approach for user interactions, which imposes a language/communication preference without indicating that the user may choose another language or locale. Under the policy, language constraints should be optional or clearly justified, and no such opt-in or justification is provided here.
This natural-language guidance presents Mandarin as the dominant language for travelers without offering user choice or acknowledging multilingual realities such as Taiwanese Hokkien, Hakka, or English support in some contexts. While not a strict forced-language instruction, it can bias users toward a single language expectation without opt-in or justification.
No suspicious patterns detected.