The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, user-provided video transcripts and marketing offer cards which are then interpreted by the model to generate localized scripts. This ingestion of untrusted data creates an attack surface where hidden instructions in a transcript could attempt to manipulate the model's output or behavior.
Ingestion points: Source transcripts and market-specific offer cards are provided as raw text inputs (SKILL.md).
Boundary markers: The skill does not define specific delimiters or instructions to the model to ignore potential commands embedded within the source transcripts.
Capability inventory: The skill possesses significant capabilities including file system access (Read/Write/Edit), media processing (ffmpeg, ffprobe), and network communication (curl).
Sanitization: There is no documented process for sanitizing or validating the contents of the transcript text before it is used in the prompt template.
[COMMAND_EXECUTION]: The skill invokes several shell-based tools to perform its primary functions, including ffmpeg for video manipulation and node/bun for running specific formatting and editing scripts.
Evidence: Uses node guide-edl.mjs and captions.mjs to re-cut and re-caption videos based on word timings.

varg-ad-localize