creative
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and analyzes live advertisements from the Meta Ad Library using Firecrawl. While intended for research, this process creates an ingestion point for untrusted data. Maliciously crafted ad text could theoretically attempt to influence the agent's generation instructions when it summarizes the research for campaign planning.\n
- Ingestion points:
scripts/ad_library.pypulls data from the Meta Ad Library into the local research directory;modes/ad-creative.mdinstructs the agent to read and log this data.\n - Boundary markers: There are no specific delimiters or instructions to ignore embedded commands within the retrieved research data.\n
- Capability inventory: The skill possesses extensive capabilities including file system access, network communication with multiple AI APIs, and execution of local scripts and CLI tools like
ffmpegandnpx.\n - Sanitization: The agent is instructed to 'study the pattern' and 'never lift another brand's words', which acts as a manual check, but there is no technical sanitization of the scraped content.\n- [EXTERNAL_DOWNLOADS]: The skill downloads model configurations and media files from reputable services including Replicate, Firecrawl, and the Meta Ad Library. These downloads are intrinsic to the skill's purpose of creative asset generation and competitor analysis.\n- [COMMAND_EXECUTION]: The skill utilizes local Python and Shell scripts to manage model verification and research tasks. It also leverages standard creative tools such as
ffmpegand thehyperframesCLI (vianpx) to process and render video assets. These commands are executed within a controlled local context as part of the intended workflow.
Audit Metadata