The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/imiva_mcp.py executes the vendor's MCP command-line interface using subprocess.Popen. The command is constructed using a hardcoded list of arguments without shell interpolation, which is a safe practice to prevent command injection.
[EXTERNAL_DOWNLOADS]: The skill uses npx to dynamically fetch and execute the official package @infimind/ecom-content-cli from the NPM registry. This is a standard and expected deployment method for the skill's core image-processing functionality.
[CREDENTIALS_SAFE]: The skill requires an MCP_TOKEN for authentication but follows best practices by directing the user to store it in a local environment variable rather than hardcoding it. The scripts access this token securely from the environment.
[SAFE]: The skill handles untrusted user input (prompts and image URLs) to perform its e-commerce tasks.
Ingestion points: Arguments to the create_smart_refine_task tool in scripts/imiva_mcp.py.
Boundary markers: Prompt templates are provided in the documentation to guide the agent, though explicit delimiters for user data are not enforced.
Capability inventory: The skill is limited to triggering image generation and task management via the vendor's API; it does not have access to sensitive system files or administrative privileges.
Sanitization: User inputs are serialized into JSON before being passed to the tool, ensuring they are not misinterpreted as shell commands or control characters.

ai-ecommerce-expert-seedream-5-lite-ecommerce-image