restaurant-booking
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements robust privacy controls for sensitive data. It instructs the agent to mask phone numbers, verify authorization before sharing health-related dietary info, and use secure sign-in sessions for payments rather than collecting new credentials.
- [INDIRECT_PROMPT_INJECTION]: The skill has a functional surface for processing untrusted data from external restaurant providers and booking emails.
- Ingestion points: Booking provider pages, confirmation emails, and user-provided notes mentioned in SKILL.md.
- Boundary markers: The instructions mandate adherence to a
<confirmation_policy>and require explicit user approval before final actions. - Capability inventory: The skill utilizes
$orbit:email, the Wallet skill, and automated browser navigation for checkout (SKILL.md). - Sanitization: The agent is directed to surface hidden terms, fees, and restrictions to the user before submission, and to verify all confirmation statuses independently.
Audit Metadata