The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/imiva_mcp.py uses npx -y @infimind/ecom-content-cli@latest to download and run an external Node.js package at runtime. This allows for the execution of code fetched from the public NPM registry during skill operation.
[COMMAND_EXECUTION]: The helper script scripts/imiva_mcp.py utilizes the subprocess.Popen module to launch system commands. While it uses list-based arguments to mitigate shell injection, it still allows the agent to execute underlying system tools like npx.
[EXTERNAL_DOWNLOADS]: The skill's functionality depends on downloading an NPM package and communicating with the external API at https://imiva.ecpro.com. This involves fetching code and data from remote sources.
[DATA_EXPOSURE]: The script scripts/imiva_mcp.py includes a --args-file option that reads the contents of a local file using Path.read_text(). If an untrusted prompt specifies a path to a sensitive file (e.g., credentials or configuration), the content could be read and processed, though it must be valid JSON to be fully parsed.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by processing untrusted product data and descriptions.
Ingestion points: Data enters the system via the productImages and extraDescription parameters in the create_detail_page_task tool.
Boundary markers: Absent; user-provided strings are passed directly to the tool without being wrapped in delimiters or safety instructions.
Capability inventory: The skill has the capability to execute subprocesses and perform network requests via the IMIVA MCP package.
Sanitization: Validation is limited to ensuring the input is a valid JSON object; no filtering or escaping of the actual string content is performed.

ai-ecommerce-expert-tmall-ecommerce-content