The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents how to build systems that process untrusted external data through RAG and Tool Calling architectures.
Ingestion points: Documentation in references/rag-advisors.md covers reading external content via PagePdfDocumentReader, TikaDocumentReader, and JsoupDocumentReader.
Boundary markers: The guide discusses system prompt setup but does not detail specific data sanitization or delimiting techniques for processed documents.
Capability inventory: The described patterns include network communication with LLM providers, local file system interactions for vector storage, and tool execution capabilities.
Sanitization: The skill proactively warns that model-generated tool calls should be treated as untrusted input and recommends validating all arguments (SKILL.md).
[COMMAND_EXECUTION]: The documentation provides examples of CLI-based configuration and service management.
It includes configuration steps for MCP servers that use npx to run the @anthropic/mcp-server-brave-search package (references/providers.md).
It outlines the process for enabling the Docker Model Runner service using the docker command line interface (references/providers.md).

spring-ai