The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill implements a dynamic code execution pattern in reference/medusa-exec.md. The executeCode function writes agent-generated TypeScript to a temporary file on the server's disk and subsequently loads it using the Node.js require() function. This allows for the execution of arbitrary code in the context of the Medusa server process.
[COMMAND_EXECUTION]: The 'MedusaExec' tool is explicitly designed to allow the AI agent to write scripts that interact with the Medusa server's internal services and database to perform queries and mutations.
[DATA_EXFILTRATION]: Because the agent has full access to the Medusa Dependency Injection container (MedusaContainer) via the code execution tool, it can access any data managed by the platform. If the agent's logic is manipulated, this capability could be abused to exfiltrate sensitive store data.
[PROMPT_INJECTION]: The skill architecture is vulnerable to indirect prompt injection as it processes data from the store to make decisions.
Ingestion points: External data enters the agent context through query.graph() results in reference/medusa-exec.md and user-provided messages in reference/api-route.md.
Boundary markers: The instructions do not define strict delimiters or 'ignore' instructions for the data being processed by the agent.
Capability inventory: The agent has the capability to write files and execute arbitrary TypeScript code via executeCode in reference/medusa-exec.md.
Sanitization: There is no implemented sanitization or validation of the agent-generated code before it is written to disk and executed by the server process.

creating-agents-in-medusa