The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text and media URLs to generate a JSON request payload. This represents a potential surface for indirect prompt injection if the downstream model or processing pipeline is vulnerable to malicious content embedded in the inputs.
Ingestion points: scripts/prepare_multimodal_embedding_request.py accepts input via command-line arguments (--text, --image, --video).
Boundary markers: None; inputs are placed directly into the JSON structure.
Capability inventory: The script writes the resulting payload to the local filesystem using pathlib.
Sanitization: No sanitization or validation is performed on the input content before payload generation.
[COMMAND_EXECUTION]: The skill documentation provides shell commands for environment setup, code validation, and script execution.
Evidence: SKILL.md contains bash commands for creating output directories, compiling Python scripts with py_compile, and executing the payload preparation script.
[EXTERNAL_DOWNLOADS]: The skill includes references to official Alibaba Cloud documentation and model studio resources.
Evidence: references/sources.md contains links to help.aliyun.com for model documentation and usage guidelines.

aliyun-qwen-multimodal-embedding