nano-banana-image-ad
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [UNVERIFIABLE_DEPENDENCIES]: Fetches markdown documentation and prompt libraries from the official novoads/claude-code-ads GitHub repository.
- [DATA_EXPOSURE]: Accesses and transmits local image files to api.novoads.ai for processing, which is required for the skill's functionality.
- [INDIRECT_PROMPT_INJECTION]: Ingests untrusted user prompts and reference images. Ingestion points: User creative direction and local image paths provided to scripts/generate_image.py. Boundary markers: Automatically appends safety and edge-safe suffixes to the final prompt. Capability inventory: Performs network requests via standard libraries and writes generated assets to the local filesystem. Sanitization: Includes a prompt rewrite phase and enforces mandatory safety suffixes to mitigate adversarial inputs.
Audit Metadata