The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes examples that fetch and execute the weather MCP server from the official Anthropic scope on npm using npx.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-supplied data into Large Language Model (LLM) prompts through the SAP Orchestration Service and standard SDK clients.
Ingestion points: Untrusted data enters the agent context via placeholderValues in the OrchestrationClient and within SAP Cloud Application Programming (CAP) event handlers.
Boundary markers: The SDK uses template syntax (e.g., {{?question}}) for interpolation, though explicit boundary delimiters in prompt examples are not strictly enforced.
Capability inventory: The skill enables capabilities such as network requests to SAP AI Core and tool-calling via LangChain and Spring AI.
Sanitization: The documentation provides robust implementation patterns for security mitigations, including Azure Content Safety filtering, Llama Guard 3 configuration, and SAP Data Privacy Integration (DPI) for automated data masking and anonymization.
[SAFE]: The skill recommends secure credential management practices, such as using VCAP_SERVICES in BTP runtimes and AICORE_SERVICE_KEY environment variables for local development, rather than hardcoding sensitive information. All external package references target official SAP registries and well-known, trusted AI frameworks.

sap-cloud-sdk-ai