The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill's primary function involves ingesting untrusted data such as article titles and summaries to generate image prompts (see references/workflow/prompt-template.md).
Ingestion points: Article title and Content summary fields in the prompt template.
Boundary markers: Absent; untrusted content is interpolated directly after labels.
Capability inventory: The resulting prompts are used to call image generation APIs which can consume tokens or produce inappropriate content.
Sanitization: None detected. An attacker could provide a malicious title designed to override the visual instructions.
[Data Exposure & Exfiltration] (LOW): The skill includes documentation and Python code (references/workflow/qwen-api.md) that performs network operations to dashscope.aliyuncs.com. While this is the intended functionality for the Qwen image generator, this domain is not on the trusted organization list. The code also utilizes the DASHSCOPE_API_KEY environment variable for authentication.
[Dynamic Execution] (SAFE): The Python code provided for image synthesis and resizing (using PIL) uses standard, safe libraries and does not employ eval(), exec(), or other dangerous dynamic execution patterns.

cover-image