AI RAG Defense

Token reduction, retrieval fidelity, and prompt-injection defense layers for the Context Engine. Covers a Summarizer agent for proactive context reduction, micro-context engineering (prompt design inside agents), high-fidelity RAG with source-metadata citations, layered untrusted-context defenses against prompt injection, and grounded-reasoning validation that prevents hallucination. Use this skill when: - Reducing token costs by adding a Summarizer agent - Designing system prompts inside specialist agents (micro-context) - Adding citations and source metadata to RAG outputs - Defending agents against prompt injection / data poisoning - Validating agent reliability acros…