The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and use "the buyer and their exact words" sourced from untrusted external locations such as reviews, social media comments, and Reddit.
Ingestion points: Input section in SKILL.md requiring "exact words about the problem (reviews, comments, Reddit)".
Boundary markers: Absent. There are no instructions to wrap these inputs in delimiters or warnings to ignore potential instructions embedded within the user-provided text.
Capability inventory: The agent has access to Bash(curl:*), Bash(ffmpeg:*), Write, Edit, and specialized AI model tools (eleven_v4, kling_v4, etc.).
Sanitization: Absent. While the skill mandates checking hooks for advertising policy compliance, it lacks technical sanitization or escaping to prevent the agent from following instructions found in the raw review text.
[COMMAND_EXECUTION]: The skill utilizes powerful shell-based tools to perform its tasks.
Evidence: The allowed-tools section explicitly permits Bash(curl:*) and Bash(ffmpeg:*).
Usage: curl is intended for communication with the vendor's API and MCP services, while ffmpeg is used via a local script scripts/hook-variants.sh to process video files.
[EXTERNAL_DOWNLOADS]: The skill connects to and references external resources for its core functionality.
Evidence: URLs pointing to https://varg.ai and https://mcp.varg.ai/mcp are included in the metadata and compatibility sections.
Purpose: These connections are used to estimate costs and interact with the vendor's Model Context Protocol (MCP) server. These resources are owned by the skill's author and are integral to the intended workflow.

product-video-hooks