ngs-amazon-image-studio
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted inputs from user-supplied product business cards (
references/business-card.md) and scraped data from external competitor listings (references/competitor-benchmark.md). This information is interpolated into prompts for image generation tools. - Ingestion points: User-provided fields (features, selling points) and content retrieved from Amazon URLs (ASINs/links) specified in the business card.
- Boundary markers: The skill employs 'Product Fact Locks' (Step 4) and 'Meta-Prompts' (Step 8) to maintain facts, though it does not use strict delimiters to separate external data from system instructions during all generation steps.
- Capability inventory: The agent has the capability to call image generation tools and execute the local
scripts/generate_image.pyscript with dynamically generated prompt content. - Sanitization: Employs 'Negative Word' lists and task-specific constraints within meta-prompts to prevent the generation of unauthorized content.
- [COMMAND_EXECUTION]: The skill executes a local Python script,
scripts/generate_image.py, to handle image generation tasks when native tools are unavailable. The script is invoked with arguments (such as the prompt and output path) determined by the agent during its reasoning process. - Evidence: Step 7 of
SKILL.mdand thereferences/image-backends.mdfile describe the invocation of this script using a command-line interface.
Audit Metadata