The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill integrates data from 13 different external marketplace connectors (e.g., Wildberries, Ozon, AliExpress, Taobao), creating a broad surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through the output of 39 namespaced tools (like wb_*, ozon_*) which fetch real-time product data and descriptions from third-party websites.
Boundary markers: The skill instructions do not specify any delimiters or safety prompts to prevent the agent from following instructions potentially embedded in product titles or descriptions.
Capability inventory: The agent has access to the marketplace-mcp CLI, which can perform filesystem writes and print configuration data.
Sanitization: No sanitization or escaping mechanisms are described for the external content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill defines an operator CLI tool, marketplace-mcp, which performs system-level operations.
The doctor command supports a --status-file argument, allowing the agent to write a JSON report to a file path, which could potentially be abused to overwrite files if the path is manipulated.
The install command generates JSON configuration strings that include filesystem paths of the current checkout.

marketplace