aiui-cloud-integration
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data that may originate from untrusted sources, creating a potential surface for indirect injection attacks where malicious content could influence agent behavior or cause unintended navigation on the device.
- Ingestion points: Data fields such as
content,customData, andtool.parametersinreferences/messaging.mdandreferences/notifications.mdare ingested from the agent context to be sent to external services. - Boundary markers: None are explicitly defined in the prompt templates, though the documentation includes warnings to validate this data.
- Capability inventory: The skill can send notifications to physical devices (
sendNotification) and store messages in a cloud cache (saveTemporaryMessage). - Sanitization: The documentation explicitly instructs users to "Validate path, customData, and content before using them for navigation or rendering" and to "Redact credentials and sensitive payloads from diagnostics," though the code implementation of this sanitization is left to the user.
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install an external package from the vendor's namespace.
- Evidence:
npm install @yodaos-pkg/cloud-integrationinSKILL.md. - [COMMAND_EXECUTION]: The skill provides
curlcommand examples to interact with remote API endpoints. - Evidence: Multiple
curlexamples inreferences/account-services.md,references/messaging.md, andreferences/notifications.mddemonstrate how to interact withrcs.rokid.comand other configured endpoints.
Audit Metadata