The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/imiva_mcp.py uses subprocess.Popen to invoke the npx command runner. This is used to launch the Model Context Protocol (MCP) server that handles the e-commerce content tasks.
[REMOTE_CODE_EXECUTION]: The skill utilizes npx -y to download and execute the @infimind/ecom-content-cli package from the official NPM registry. This package is the official command-line tool for the IMIVA platform mentioned in the skill documentation.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes external product images, reference videos, and user-provided descriptions.
Ingestion points: Parameters such as referenceVideo and productImages used in the create_viral_video_copy_task tool.
Boundary markers: The skill documentation enforces a staged workflow where the AI must return the video formula and script for manual human confirmation before final generation.
Capability inventory: The skill interacts with the file system and network through the IMIVA MCP tool executed in scripts/imiva_mcp.py.
Sanitization: All arguments passed to the underlying tool are serialized using standard JSON formatting to prevent command injection at the application level.
[SAFE]: The skill correctly instructs users to store sensitive MCP_TOKEN credentials in environment variables rather than hardcoding them within the skill files or scripts.

ai-ecommerce-expert-viral-video-staged-plan