coupang-product-search
Pass
Audited by Gen Agent Trust Hub on Apr 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill communicates with an external endpoint hosted on Hugging Face Spaces (
yuju777-coupang-mcp.hf.space) to proxy requests to Coupang and Danawa. Hugging Face is a well-known hosting platform, and the communication is required for the skill's primary function. - [COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using
curl,grep,awk, andtrto initialize a session and retrieve a session ID from the remote server's response. While these are standard text-processing utilities, they are being applied to data received from an external network source. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
- Ingestion points: Product names, descriptions, and option details are fetched from the
yuju777-coupang-mcp.hf.spaceendpoint. - Boundary markers: The instructions do not define clear delimiters or warnings to ignore instructions embedded within the product data.
- Capability inventory: The agent is instructed to display the retrieved data and generate comparison tables, which could potentially include instructions that influence the agent's next steps if the product data is malicious.
- Sanitization: There is no explicit sanitization or validation of the external content before it is processed by the agent.
Audit Metadata