The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates communication with official xAI API endpoints (api.x.ai) and media delivery servers (vidgen.x.ai) for video generation and artifact retrieval. These interactions are strictly governed by a host-policy allowlist and DNS-pinning logic within the provided script to prevent SSRF and unauthorized data access.
[COMMAND_EXECUTION]: The included Python utility xai_video_guard.py uses the subprocess module to execute ffprobe and ffmpeg. These tools are utilized solely for legitimate media validation, such as verifying file signatures, probing duration/resolution for accurate pricing, and ensuring full artifact decoding before human review.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external user prompts and media files (images/videos) which are interpolated into API requests. To mitigate injection risks, the skill implements a comprehensive security framework:
Ingestion points: Untrusted data enters the context through user-defined prompt strings and local/remote media files referenced by URLs or file identifiers.
Boundary markers: The instructions mandate the use of explicit creative beats and causal arcs in prompts to maintain control, while the script uses SHA-256 digests to lock the intent before execution.
Capability inventory: The skill manages network POST/GET operations, local file system writes for artifacts, and subprocess calls for media verification.
Sanitization: The script performs multi-layer validation, including signature checks (JPEG/PNG/WebP/MP4), duration/resolution probing, and cryptographic hashing of all source inputs to ensure integrity and policy compliance.

xai-grok-imagine-video