The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided text prompts and external image/video URLs, which are passed to the Kling Video model via an external API. This creates a surface for indirect prompt injection where adversarial content in the input could attempt to influence the model's behavior or the agent's processing of the output.
Ingestion points: The scripts/generate_video.sh script accepts prompt text (--prompt) and reference URLs (--first-frame, --reference-image, etc.) as input arguments.
Boundary markers: No specific delimiters or safety instructions are added to the prompt payload to separate user content from system instructions.
Capability inventory: The skill uses curl for API interaction and python3 for data processing, enabling network requests and file reads.
Sanitization: Input values are safely encoded into JSON using Python's json module, protecting against structural injection, but the semantic content of the prompts is not filtered.
[COMMAND_EXECUTION]: The skill utilizes curl for network communication with the Frevana API and for downloading generated video files. It also uses python3 for JSON manipulation and image data encoding. These operations are core to the skill's functionality and implemented using standard practices.
Evidence: Implementation in scripts/generate_video.sh utilizes these tools for creating jobs, polling status, and downloading results. The unit tests also utilize subprocess.run in a safe, list-based manner to verify script behavior.

kling-v3.0-pro