The Agent Skills Directory

[COMMAND_EXECUTION]: The generate_video.sh script executes curl for network requests and python3 for processing JSON payloads and encoding image data. The test files similarly use subprocess.run to execute the shell scripts for verification.
[DYNAMIC_EXECUTION]: The bash script utilizes Python heredocs to dynamically construct JSON payloads and perform base64 encoding of image files. This execution is limited to data transformation using local environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input via the --prompt argument and processed image files without explicit boundary markers or sanitization before passing them to the video generation model.
Ingestion points: User-provided strings via --prompt and local/remote files via --first-frame, --last-frame, and --reference-image in generate_video.sh.
Boundary markers: None identified in the prompt construction logic; user input is directly interpolated into the API payload.
Capability inventory: The skill uses curl to make network requests, python3 for data processing, and basic file system operations such as mkdir and cp.
Sanitization: The script performs format validation on parameters like --duration, --aspect-ratio, and --size, but does not sanitize the contents of the prompt or images.

minimax-h3