ad-creative

Pass

Audited by Gen Agent Trust Hub on Oct 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references and provides installation instructions for various third-party tools and libraries needed for advanced video and audio production, such as gooseworks, remotion, and voicebox (hosted on GitHub). It also encourages the use of multiple external AI APIs for image and video generation.\n- [COMMAND_EXECUTION]: The documentation includes examples of using local Node.js CLI tools to interact with advertising platform APIs (e.g., google-ads.js) to pull performance metrics and reports.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process customer-generated content like reviews and ad comments from the inputs/ folder to ensure generated ads are grounded in real customer language.\n
  • Ingestion points: Reads customer reviews and comments from inputs/reviews/ and inputs/comments/.\n
  • Boundary markers: The instructions explicitly mandate citing specific sources and forbid inventing claims or testimonials, reducing the risk of unintended instruction following.\n
  • Capability inventory: The agent has access to campaign management tools for Google, Meta, and LinkedIn Ads, and can write output files to dated batch folders.\n
  • Sanitization: The skill provides specific instructions for escaping HTML characters (< to \\u003c) when populating the JSON data block in its review templates to prevent script injection.\n- [DYNAMIC_EXECUTION]: The creative-review-template.html asset uses a client-side JavaScript engine to parse a JSON data block and dynamically generate in-feed ad mockups and storyboards for stakeholder review.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 2, 2026, 09:00 PM