ai-builder

Pass

Audited by Gen Agent Trust Hub on Apr 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill consists entirely of markdown documentation and pseudo-code examples for Microsoft Power Platform configurations. It contains no executable scripts, shell commands, or binary files that could pose a risk to the host environment.
  • [PROMPT_INJECTION]: The skill documents methods for processing untrusted external data (e.g., customer feedback, support tickets) using GPT models, creating an indirect prompt injection surface. However, it explicitly instructs users to test for 'injection attempts' and provides robust prompt engineering patterns (Role + Task + Format) to mitigate these risks.
  • Ingestion points: Data input variables such as {{CustomerFeedback}}, {{TicketDescription}}, and {{OrderEmail}} in gpt-prompts.md and document-processing.md.
  • Boundary markers: Recommends the use of system instructions and explicit output formatting requirements (e.g., JSON) to constrain the model's behavior.
  • Capability inventory: Described actions include writing to Dataverse tables, sending Teams notifications, and sending automated emails via Power Automate.
  • Sanitization: Advises on the use of Parse JSON and IfError() logic in Power Automate to validate and handle model outputs safely.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 10, 2026, 02:03 AM
Security Audit — agent-trust-hub — ai-builder