lyt-image-patterns
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill defines a rigorous workflow for analyzing product image samples. It includes specific 'Causal Guardrails' and 'Prohibited Items' sections that prevent the agent from providing misleading advice or overestimating the impact of visual changes on sales performance.
- [EXTERNAL_DOWNLOADS]: The skill handles external input in the form of product links from established platforms such as 1688 and TikTok Shop. This is a primary function of the skill intended for competitive analysis and does not involve the execution of remote code.
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from untrusted external URLs, creating a standard surface for indirect prompt injection.
- Ingestion points: External links from e-commerce sites and competitive shop samples (SKILL.md).
- Boundary markers: The skill mandates the use of specific output structures, 'Sample Boundaries' (样本边界), and validation steps to isolate sample data from the analysis logic.
- Capability inventory: Analysis is restricted to visual and text processing within the agent's context; no dangerous tools such as shell execution or file-system writing are invoked.
- Sanitization: Instructions require the agent to clean samples and categorize them into structured tables, reducing the likelihood of processing embedded malicious instructions as commands.
Audit Metadata