The Agent Skills Directory

[DATA_EXFILTRATION]: The skill describes procedures for transmitting local media files (images and videos) to the Kling AI API. It correctly identifies this as a potential data movement path and instructs the agent to ask for user confirmation before uploading local files, preventing silent exfiltration of local media.
[INDIRECT_PROMPT_INJECTION]: The skill processes external inputs which are interpolated into API calls, creating a potential attack surface.
Ingestion points: User-provided prompt text and external media URLs or file paths provided at runtime.
Boundary markers: Instructions recommend using stable prompt placeholders (e.g., <<<image_1>>>) to bind specific references and validate image existence before prompt insertion.
Capability inventory: The skill performs network POST requests to Kling AI endpoints, reads local files to convert them to Base64 strings, and downloads generated media items to local storage.
Sanitization: Includes explicit validation logic for callback_url parameters, ensuring only absolute http or https URLs are processed.
[EXTERNAL_DOWNLOADS]: The skill is configured to fetch generated audio and video assets from official Kling AI endpoints (e.g., api-singapore.klingai.com) after task completion.
[SAFE]: No malicious patterns such as obfuscation, persistence, or privilege escalation were detected. The skill uses standard environment variable configurations for authentication and implements conservative cost estimation and error handling.

kling-official