The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill describes a 'Dynamic Hubs' feature (Section 11) that allows method definition and dispatch at runtime. It specifically mentions that these hubs can be defined from 'downloaded web apps or scripts'. This creates a risk of executing unverified or malicious logic if the hub handler implementation does not apply strict constraints on the methods and arguments provided by external sources.
[INDIRECT_PROMPT_INJECTION]: The library is designed to ingest and process data from external network sources (BLE, Wi-Fi, Nearby, Multipeer). This creates a surface for indirect prompt injection if an AI agent were to use this library to act upon messages from untrusted peers. Ingestion points: Network data enters the system through 'HubConnection', 'DynamicHubCall.Arguments', and 'IBleHubSerializer' implementations. Boundary markers: While the protocol uses structured framing, the skill does not instruct developers to use security delimiters for the payload content. Capability inventory: The library includes capabilities for remote method invocation and file system access via file transfers. Sanitization: The instructions lack guidance on sanitizing incoming data before it is used.
[EXTERNAL_DOWNLOADS]: The documentation references multiple external dependencies, including the 'Shiny.UniversalHubs' suite of packages and Google Nearby Connections. These resources are consistent with the vendor's own ecosystem and established services.
[COMMAND_EXECUTION]: Section 12 mentions a native build script ('native/ShinyNearbyBridge/build.sh'). Although noted as being for package maintainers, it represents an executable script path within the project structure.

shiny-universal-hubs