The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data provided by the user, specifically text prompts and images (starting frames and reference images), which are expanded by the agent and then sent to external AI generation APIs such as Gemini, xAI, and Replicate. The generated output is written to the local filesystem.\n
Ingestion points: Positional command-line arguments and file path flags (--input, --ref, --last-frame) processed by scripts/generate.ts.\n
Boundary markers: Absent. The instructions provide guidance for prompt rewriting but do not specify delimiters or instructions for the agent to ignore potentially malicious content embedded in the user input.\n
Capability inventory: The skill can perform network requests to multiple AI providers and write .mp4 and .png files to the filesystem.\n
Sanitization: The script does not perform technical sanitization (such as character filtering or escaping) on the expanded prompt string before it is passed to the generation SDKs.\n- [DYNAMIC_EXECUTION]: The script scripts/generate.ts uses dynamic import() calls to load internal modules based on paths calculated at runtime.\n
Runtime loading: A resolution helper function locates the plugin's installation root by reading platform-specific configuration files (e.g., ~/.claude/plugins/installed_plugins.json) and scanning local cache directories, then dynamically imports logic from files like utils.ts, shared.ts, and providers/xai.ts using the resolved path.

generate-video