ad-creative

Warn

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends several external resources and third-party tools for creative production, including a GitHub repository from an individual developer (github.com/jamiepine/voicebox.git) and the Remotion video framework via NPM.- [REMOTE_CODE_EXECUTION]: The instructions for setting up recommended tools involve downloading and executing code from the internet, specifically through the use of npx create-video@latest and running make setup on a cloned repository.- [COMMAND_EXECUTION]: Documentation for the skill includes various shell command templates for interacting with APIs via curl, processing video with ffmpeg, and executing local scripts using node.- [PROMPT_INJECTION]: The skill is designed to ingest and analyze external data such as performance metrics and marketing context, which serves as a vector for indirect prompt injection.
  • Ingestion points: Performance data provided via CSV, text paste, or API output, and context files like product-marketing-context.md.
  • Boundary markers: There are no defined delimiters or instructions provided to the agent to separate data from potentially malicious instructions within the inputs.
  • Capability inventory: The skill is configured to use high-privilege capabilities including shell command execution (ffmpeg, npx, git), network operations (various AI APIs), and file system access.
  • Sanitization: The instructions do not specify any validation or sanitization requirements for the external data being processed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 7, 2026, 11:48 PM
Security Audit — agent-trust-hub — ad-creative