The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill constructs prompts and API payloads using user-supplied content, which creates a surface for indirect prompt injection.
Ingestion points: External data enters the agent context through placeholders in SKILL.md, including <lock sheet>, <scene>, and CLIP_URL.
Boundary markers: The skill suggests using structured headers such as PRODUCT:, FIXED GEOMETRY:, and FIXED SURFACE: to delineate inputs, which can help prevent the agent from misinterpreting instructions inside the data.
Capability inventory: The skill is authorized to use curl, ffmpeg, jq, and ffprobe as specified in the allowed-tools frontmatter.
Sanitization: The skill does not describe specific sanitization or escaping mechanisms for the user data before it is interpolated into the curl command or the AI prompts.
[COMMAND_EXECUTION]: The skill utilizes curl to interact with the author's API at api.varg.ai for video analysis and ffmpeg for local video compositing. These tools are used in accordance with the skill's primary function of video generation and quality control.
[EXTERNAL_DOWNLOADS]: The skill provides links to external methodology guides and examples on seedance.tv and a GitHub repository (AuraJZ/ai-product-video). These are informational references and do not represent automated execution of untrusted code.

varg-product-lock