The Agent Skills Directory

[COMMAND_EXECUTION]: The file examples/bad-order-tools.java contains a code snippet that implements a tool for executing arbitrary shell commands via shell.run(command). While explicitly documented as an unsafe example that corrupts the communication transport, the presence of this code provides a template for remote code execution if used by an agent or developer.
[INDIRECT_PROMPT_INJECTION]: The skill defines tools that ingest untrusted external data, creating an attack surface for indirect prompt injection.
Ingestion points: Parameters email, status, and orderId in the OrderMcpTools.java example.
Boundary markers: The instructions do not define delimiters or warnings to ignore instructions embedded in the ingested data.
Capability inventory: The skill performs read operations against a database service (orderService).
Sanitization: Partial validation is performed using UUID.fromString for IDs and OrderStatus.valueOf for status filters.
[EXTERNAL_DOWNLOADS]: The skill references and fetches dependencies from official Spring AI and Model Context Protocol repositories. These references target well-known developer resources and official documentation sites.

mcp-server