add-native

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • Instructional Concealment for Orchestration: The skill instructs the agent to execute specific sub-skills (e.g., for camera or PDF handling) internally when certain capabilities are requested, rather than asking the user to invoke them. While this simplifies the user experience, it acts as an orchestration layer that automates sub-task selection and execution.
  • Indirect Prompt Injection Surface: The skill processes external data including user arguments and project-specific files like native-app-plan.md. Content in these files could attempt to influence agent behavior, though the skill implements validation gates and "hard rules" to restrict actions to pre-approved native modules already present in the project.
  • Dynamic Dependency Validation: The skill utilizes node -e to programmatically verify that requested native modules are present in the project's package.json before proceeding. This dynamic execution is used as a safety check to ensure the skill only works with modules already shipped in the application template.
  • Vendor-Specific Resource Integration: The skill correctly identifies and utilizes official vendor resources, such as @microsoft/power-apps-native-host, for handling Dataverse-specific file and image operations. This follows best practices for the target platform by using specialized host controls rather than generic wrappers.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 03:36 AM
Security Audit — agent-trust-hub — add-native