The Agent Skills Directory

[SAFE]: The skill provides documentation and educational code snippets for building LLM-based applications, including RAG pipelines and vector database management. The content is technical and consistent with its stated purpose of teaching LLMOps.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates patterns for RAG and evaluation that ingest external data into LLM prompts. While these are standard architectural examples, they represent a potential attack surface for indirect prompt injection if deployed without robust sanitization.
Ingestion points: index_document (SKILL.md) ingests external content, and rag_query (SKILL.md) interpolates retrieved data directly into prompts.
Boundary markers: The examples use basic string interpolation and newlines but lack formal delimiters or safety instructions to ignore embedded commands.
Capability inventory: The snippets demonstrate local file storage via chromadb.PersistentClient and network API calls via the Anthropic client.
Sanitization: The educational snippets do not include explicit input sanitization or verification logic.

llm-ops