The Agent Skills Directory

[SAFE]: The skill consists of educational documentation and code snippets for implementing RAG patterns using standard libraries such as LangChain. All network interactions are directed toward well-known services (e.g., OpenAI, Pinecone, Weaviate) or local instances.
[PROMPT_INJECTION]: The skill describes systems that ingest external data, creating an indirect prompt injection surface. The provided examples include common mitigations such as boundary markers within prompt templates to separate retrieved context from user instructions.
Ingestion points: The DirectoryLoader reads files from the local ./docs directory.
Boundary markers: Prompt templates in the documentation use explicit headers like 'Context:' and 'Question:' to delineate data.
Capability inventory: The skill outlines patterns for LLM generation, vector database interaction, and local file reading.
Sanitization: Standard for example code, the snippets do not include complex input sanitization logic, relying instead on structural delimiters.
[CREDENTIALS_UNSAFE]: The code snippets include placeholders for API keys (e.g., api_key="your-api-key"). These are standard documentation placeholders and do not represent a credential leak.

rag-implementation