The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/imiva_mcp.py uses npx to download and execute the @infimind/ecom-content-cli@latest package during operation. This mechanism allows the execution of remote code that is not bundled with the skill, presenting a risk if the external package or registry is compromised.
[EXTERNAL_DOWNLOADS]: The skill requires the @infimind/ecom-content-cli package from the public npm registry. Using the @latest version tag means the code executed is not pinned, making the skill susceptible to supply chain attacks where a package update could introduce malicious behavior.
[COMMAND_EXECUTION]: The Python helper script uses subprocess.Popen to invoke shell commands (npx). While it passes arguments as a list to prevent basic shell injection, it still facilitates the execution of binary logic sourced from an external network package.
[DATA_EXFILTRATION]: The imiva_mcp.py script includes an --args-file parameter that reads local files into memory to parse them as JSON. This capability could be exploited by a manipulated agent to access and exfiltrate sensitive local data, provided the target files are in JSON format.
[PROMPT_INJECTION]: The skill ingests untrusted data such as reference video paths, product images, and requirement descriptions. It lacks explicit sanitization, validation, or boundary markers to prevent indirect prompt injection, where instructions embedded in the input media or text could override the agent's behavior. Evidence: The create_viral_video_copy_task tool directly accepts user-controlled media and text inputs without filtering.

ai-ecommerce-expert-viral-video-product-adaptation