skills/zoom/skills/video-sdk/windows/Gen Agent Trust Hub

video-sdk/windows

Pass

Audited by Gen Agent Trust Hub on Aug 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for handling untrusted input from session participants, which could be exploited for indirect prompt injection if the agent processes this data without boundaries.
  • Ingestion points: The skill implements callbacks for receiving external data in examples/session-join-pattern.md (chat), examples/transcription.md (transcription), and examples/command-channel.md (custom signaling).
  • Boundary markers: The code samples lack delimiters or 'ignore' instructions when passing these participant-controlled strings to the application logic.
  • Capability inventory: The skill integrates extensive networking, file system access (for recording), and media control functionalities.
  • Sanitization: No sanitization, escaping, or validation of participant-provided strings is shown in the provided implementation examples.
  • [DYNAMIC_EXECUTION]: The skill demonstrates the use of Foreign Function Interface (FFI) to invoke native Windows system libraries.
  • Evidence: examples/dotnet-winforms/README.md utilizes [DllImport("gdi32.dll")] for native memory management during image processing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 26, 2026, 09:56 PM
Security Audit — agent-trust-hub — video-sdk/windows