The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill relies on an external Node.js package (@shopme/visual-product-search-mcp) executed via npx. This package is not from a trusted organization listed in the security policy, posing a risk of executing unverified code from the public npm registry.
COMMAND_EXECUTION (LOW): The recommended MCP configuration uses npx, which downloads and executes the latest version of the specified tool at runtime. While standard for many tools, it bypasses static version pinning and local audits.
PROMPT_INJECTION (LOW): (Category 8
Indirect Prompt Injection) The extract_product_from_image tool processes external image URLs to extract product names and detected text. This represents an attack surface where malicious text embedded in images could influence the agent's behavior.
Ingestion points: imageUrl in extract_product_from_image and search_by_image tools.
Boundary markers: Absent. The skill does not instruct the agent to ignore instructions found within the processed images.
Capability inventory: The skill returns structured text data which influences the agent's downstream reasoning and search parameters.
Sanitization: None mentioned; the AI vision model's output is passed directly to the agent context.
CREDENTIALS_UNSAFE (LOW): The skill requires five distinct API secrets (Taobao App Secret/Token, AliExpress DS Secret, Gemini API Key). While the skill doesn't hardcode them, the accumulation of high-value credentials in an unverified tool increases the risk of exfiltration if the MCP server is compromised.

visual-product-search