The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The skill explicitly directs the agent to execute a local script located at /Users/alchain/Documents/写作/tools/upload_image.py using the system shell (python3).
Evidence: Multiple instructions under '图床上传流程' and 'Step 3' command the agent to run python3 /Users/alchain/Documents/写作/tools/upload_image.py <图片URL或本地路径>.
Risk: If an attacker can influence the input to this skill (e.g., via a malicious article), they might attempt shell injection or trick the agent into executing the script with unintended local file paths.
CREDENTIALS_UNSAFE (MEDIUM): The skill documentation reveals that sensitive credentials (ImgBB API key) are stored in the user's shell configuration file.
Evidence: The '技术依赖' section states: 'API配置: ImgBB API key（已配置在 ~/.zshrc）'.
Risk: This encourages the agent to interact with or rely on environment variables from sensitive system files like ~/.zshrc, which could lead to accidental exposure of the file's full contents if the agent is asked to debug or read its environment.
DATA_EXFILTRATION (LOW): The core functionality involves sending data to an external third-party service (ImgBB).
Evidence: The skill is designed to take local or remote images and upload them to i.ibb.co.
Risk: While this is the stated purpose, the lack of sanitization on the file path input means a malicious prompt could potentially exfiltrate sensitive documents by passing their paths to the upload script instead of image files.

image-illustration-upload