shiny-contactstore
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents an integration for AI agents through the
Shiny.Contacts.Extensions.AIpackage, which exposes device contact data to LLM functions. This creates a surface for indirect prompt injection if contact data contains malicious instructions. - Ingestion points: Device contacts (names, notes, organizations) ingested via
IContactStoreinSKILL.md. - Boundary markers: The skill emphasizes that developers must explicitly opt-in to specific capabilities (Read/Write) when exposing tools to the agent.
- Capability inventory: The integration allows agents to search, create, update, and delete contacts (identified in the AI Tool Integration section).
- Sanitization: The documentation suggests model-level confirmation for destructive actions but does not specify input sanitization for contact fields.
Audit Metadata