The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs runtime installation of a third-party package.
Evidence: scripts/imiva_mcp.py and SKILL.md instruct the agent to run npx -y @infimind/ecom-content-cli@latest, which downloads and executes code from the NPM registry every time the skill is initialized.
[COMMAND_EXECUTION]: The skill uses Python's subprocess module to run shell commands.
Evidence: scripts/imiva_mcp.py uses subprocess.Popen to execute the npx command. While the command list is currently restricted to the package name, it expands the attack surface of the environment.
[REMOTE_CODE_EXECUTION]: The skill creates a pattern for remote code execution by combining external downloads with direct system execution.
Evidence: The @infimind/ecom-content-cli package is fetched from a remote source and executed on the local machine with the user's permissions.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external inputs through its core tools.
Ingestion points: The create_product_main_image_task tool accepts productImages (URLs or paths), usageScenario, and coreSellingPoints as input.
Boundary markers: None identified; user-provided text and external image metadata are interpolated into the workflow without explicit isolation.
Capability inventory: The skill has the ability to execute shell commands (via npx) and make network requests (via the IMIVA API).
Sanitization: No input sanitization or validation is present in scripts/imiva_mcp.py before passing arguments to the MCP bridge.

ai-ecommerce-expert-home-furniture-ecommerce-content