The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts/generate_image.py script accesses sensitive configuration and credential files at ~/.alibabacloud/credentials and .env to retrieve API keys. This is standard behavior for an Alibaba Cloud integration but involves access to sensitive data.
[DATA_EXFILTRATION]: The resolve_reference_image function in scripts/generate_image.py reads the binary content of any local file path provided in the reference_image parameter. Since this content is then transmitted to the external DashScope API, it creates a risk of exfiltrating sensitive local files (e.g., SSH keys or environment configs) if an attacker can control the input path.
[COMMAND_EXECUTION]: The skill requires executing local Python scripts (scripts/generate_image.py) and shell commands (curl, python -m pip install) to perform its tasks.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8).
Ingestion points: User-provided prompt, negative_prompt, and reference_image paths within the JSON request processed by scripts/generate_image.py.
Boundary markers: None detected in the script or instructions to isolate untrusted user inputs from system instructions.
Capability inventory: File system read (Path.read_bytes), network transmission via DashScope SDK (ImageGeneration.call), and file system write (Path.write_bytes via download_image).
Sanitization: No sanitization or validation is performed on the prompt text or the file paths provided for reference images.

alicloud-ai-image-qwen-image