The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (MEDIUM): The scripts generate.sh, poll.sh, and submit.sh all use curl -k (or --insecure) to communicate with the DashScope API and download media files. This bypasses SSL certificate validation, making the communication vulnerable to interception and modification by attackers.
[COMMAND_EXECUTION] (MEDIUM): In submit.sh, the JSON request body is constructed by manually concatenating shell variables like $PROMPT and $AUDIO_URL. This lacks proper escaping, allowing a malformed or malicious prompt to break the JSON structure or potentially influence the API request in unintended ways.
[CREDENTIALS_UNSAFE] (LOW): The skill requires the DASHSCOPE_API_KEY environment variable. While the key is not hardcoded, the scripts' use of insecure curl calls means this secret is transmitted without verified encryption, increasing the risk of credential theft.
[PROMPT_INJECTION] (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) due to its data handling practices.
Ingestion points: User-provided strings in --prompt and --audio-url arguments.
Boundary markers: None; inputs are directly embedded into JSON strings.
Capability inventory: Network requests and file-system writes via curl and mkdir.
Sanitization: No escaping or validation is performed on the input variables before they are used in commands or payloads.

qwen-video