The Agent Skills Directory

[DATA_EXFILTRATION]: The scripts scripts/generate_video.py and scripts/generate_dancing_video.py are designed to read sensitive authentication tokens from the local Alibaba Cloud credentials file at ~/.alibabacloud/credentials. This facilitates interaction with the cloud provider but exposes local system secrets to the skill's execution context.
[EXTERNAL_DOWNLOADS]: The skill uses urllib.request to download generated video and image files from remote URLs. These URLs are retrieved dynamically from the Alibaba Cloud DashScope API responses after processing the user's requests.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests untrusted user text to drive AI model behavior and subsequent system actions.
Ingestion points: User-provided prompts and reference image paths are accepted in the generate_video.py and generate_dancing_video.py scripts and passed directly to the DashScope API.
Boundary markers: No delimiters or defensive instructions are implemented to prevent the model from interpreting malicious commands embedded within the user's prompt.
Capability inventory: The skill possesses network access (via urllib.request.urlopen) to fetch remote content and file system write permissions (via Path.write_bytes) to save assets locally.
Sanitization: The input is processed as standard strings without specific sanitization or filtering intended to detect or neutralize prompt injection attempts.

aliyun-wan-video