The Agent Skills Directory

[COMMAND_EXECUTION]: The skill operates primarily by executing the vg generate command-line tool to perform media uploads, status checks, and model invocations. This is the core functional design of the skill.
[EXTERNAL_DOWNLOADS]: The documentation references the vibedgames package, which is required to be installed via npm to provide the CLI tool. This is a standard dependency for the author's media processing ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an inherent attack surface for indirect prompt injection as it ingests user-provided text and media to construct prompts for third-party AI models.
Ingestion points: Processes local files (via upload) and user-supplied prompts across multiple reference files like references/realism.md.
Boundary markers: The skill advocates for structured prompt templates (e.g., SCLCAM), though it lacks explicit delimiters to isolate untrusted user input within the final prompt.
Capability inventory: The vg generate tool has the capability to read local files for upload and write results to the local filesystem using the --download flag.
Sanitization: There are no specific instructions for sanitizing or escaping user-provided data before it is interpolated into the generation commands.

media-workflow