The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is configured to download the @infimind/ecom-content-cli package from the NPM registry at runtime. Evidence is found in the scripts/imiva_mcp.py script and the references/mcp-config.example.json file, which specify the use of npx -y to fetch the package.
[REMOTE_CODE_EXECUTION]: The skill executes code downloaded from a remote source via npx -y @infimind/ecom-content-cli@latest. The use of the @latest version tag is a security risk as it allows for the execution of unpinned, potentially changing code from a third-party provider, which is a common vector for supply chain attacks.
[COMMAND_EXECUTION]: The script scripts/imiva_mcp.py uses subprocess.Popen to programmatically invoke shell commands. Specifically, it executes npx with arguments to run the ecommerce content CLI, passing environment variables like MCP_TOKEN to the subprocess.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to how it processes external data.
Ingestion points: Untrusted data enters the agent context through parameters such as usageScenario, coreSellingPoints, and productCategory in the create_product_main_image_task tool.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within these fields in the provided code.
Capability inventory: The skill has the capability to execute shell commands via subprocess and perform network operations through the MCP tool interface.
Sanitization: The script performs no sanitization or validation on the input strings beyond basic JSON encoding before passing them to the execution environment.

ai-ecommerce-expert-shopee-ecommerce-content