Security & RAG

A mental model for how prompt injection, tool access, and retrieval combine into real risk.

A hands-on intro to scoping tools, validating inputs, and wiring a first retrieval pipeline.

Sanitize and isolate untrusted retrieved content so it cannot hijack Claude's instructions or tools.

Enumerated steps for restricting tool permissions, scopes, and blast radius before deployment.

Enumerated compliance steps for redacting, retaining, and auditing PII across Claude API traffic.

Enumerated grounding techniques to keep Claude's answers tied to retrieved source documents.

Enumerated best practices for securing tool use, protecting PII, and grounding RAG answers.