The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/imiva_mcp.py uses npx -y to download and execute the @infimind/ecom-content-cli@latest package from the NPM registry every time the skill is run. Executing code that is not bundled with the skill and is not version-pinned introduces a supply chain risk where the external code could be replaced with malicious versions by the package maintainer.
[COMMAND_EXECUTION]: The skill uses the subprocess module to run local commands (npx) to facilitate communication with the IMIVA platform. While the command structure is largely static, it still constitutes the execution of a subprocess triggered by agent instructions which could be exploited if the package name or environment variables were manipulated.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests untrusted data via subjectDescriptions and productImages (in SKILL.md and scripts/imiva_mcp.py) and passes this data to downstream AI models through the MCP tool. The skill lacks boundary markers, such as XML tags, or explicit instructions to ignore embedded commands within this data. This creates a surface where adversarial content (e.g., in a product description) could influence the agent's behavior or cause it to ignore its safety constraints. The capability to perform network operations through the MCP tool increases the potential impact of such an injection.

ai-ecommerce-expert-single-reference-batch-products