The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/smoke_test_qwen_vl.py uses importlib.util to dynamically load and execute logic from a target file (analyze_image.py) using a path computed relative to the script's location.
[COMMAND_EXECUTION]: The unit test file test_analyze_image.py employs a similar pattern, using importlib.util.spec_from_file_location and spec.loader.exec_module to dynamically load the code under test from the local file system.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from the user and external images to be processed by a multimodal AI model.
Ingestion points: The prompt and image arguments in scripts/smoke_test_qwen_vl.py allow for arbitrary text and image paths/URLs to be passed to the model.
Boundary markers: There are no explicit delimiters or instructions provided in the script to separate user-provided data from the system prompt.
Capability inventory: The script executes AI model calls via mod.call_analyze, which could trigger unintended model behaviors if the input is poisoned.
Sanitization: No input validation or sanitization is performed on the prompt or image input before being sent to the model interface.

aliyun-qwen-vl-test