Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed AdsTurbo video-transformation wrapper, with expected remote media upload behavior and no evidence of hidden persistence or malicious actions.
Install only if you are comfortable sending selected videos, images, and related media URLs to AdsTurbo for processing. Avoid uploading sensitive, private, regulated, or unauthorized likeness/voice content unless you accept that the upload becomes externally accessible and subject to AdsTurbo's handling and billing terms.
The activation guidance uses broad phrases like "swap the person," "translate the video," and "make this photo move," which can match many ordinary user requests without enough gating. This can cause the agent to invoke a transformation skill when the user may have intended a different operation, increasing the chance of unintended external API use, content modification, or charges.
The Chinese trigger phrases are similarly broad and overlap with common requests, such as changing a person, translating a video, or animating a photo, without explicit scope constraints. In a multilingual agent, this raises the risk of accidental skill invocation and unintended calls to a paid third-party service based on ambiguous natural-language input.
The instructions explicitly tell users to upload local assets so they become publicly accessible URLs, but they do not warn that uploaded files may be exposed to third parties or should not contain sensitive content. This creates a real privacy and data-exposure risk because users may upload personal photos, voice samples, or videos without understanding the public accessibility implications.
The Chinese-language section repeats the same workflow of uploading local files for conversion into publicly accessible URLs, again without any explicit warning about privacy, confidentiality, or public exposure. Because this section targets the same sensitive asset types, it carries the same risk of inadvertent disclosure of personal or proprietary media.
The documentation instructs users to make video and image assets publicly accessible but does not warn that doing so can expose sensitive personal data, copyrighted media, or internal campaign materials to unintended parties. In the context of face-swapping, portrait driving, and voice/lip-sync translation, the privacy risk is heightened because the uploaded assets may contain biometric data and personally identifiable content.
The upload method reads a local file and sends its contents to a remote API, which is a safety-relevant data transmission. While the module docstring describes an HTTP client generally, this specific operation has no confirmation prompt or explicit disclosure near the upload action itself that local file contents will be transmitted off-system.
This script explicitly uploads a user-supplied local file to remote storage and the module docstring states the result is a public URL, but the CLI provides no warning, confirmation, or clear notice about data leaving the host and becoming externally accessible. That creates a real risk of accidental disclosure of sensitive local files, especially because the interface accepts arbitrary paths and presents the action as a simple utility command.
The natural-language description advertises translate as re-voicing a video in another language, which introduces a language/locale behavior. While the CLI accepts --target-lang, the file does not document any user opt-in or language policy choice around language handling beyond performing the conversion.
This code submits user-provided media URLs and related parameters to remote API endpoints via submit_and_maybe_poll, which is a network operation that can transmit user data. In this file there is no visible confirmation prompt, print/log disclosure, or warning comment/docstring describing that these inputs are sent to an external service.
