lyonzin/knowledge-rag

RAG Check First requires a local knowledge search before answering technical questions, code requests, or team-specific factual claims. Use it to ground responses in indexed docs, ADRs, runbooks, and prior work.

RAG Code Review consults related ADRs, standards, similar files, and prior incidents before commenting on a code change. Use it for PR reviews or any critique of a diff so feedback is grounded in the team’s own decisions.

RAG Evaluate Quality measures retrieval performance with MRR@5, Recall@5, Precision@5, and index health stats. Use it weekly, after reindexing, or when answer quality seems to drop.

RAG Index Decisions prompts you to turn important architectural choices, bug fixes, and team conventions into indexed documents. Use it when a conclusion should be searchable the next time the same issue comes up.

RAG Onboard Context probes the indexed knowledge base at the start of a session or after a topic shift. It checks index stats, categories, and a few sample searches so the agent knows what content is available before answering.

RAG Security First routes security questions through the local corpus before external threat-intel lookups. Use it for incident response, MITRE mapping, CVE analysis, detections, and red/blue team research grounded in local playbooks.

RAG Web Fallback forces an agent to search the local knowledge base before using external web tools. It is used when a question might be answered from indexed docs and the agent must explain why it escalated if local coverage is missing.

Rag Cite Sources requires every technical claim drawn from the local corpus to include an inline path:line or path:section citation. Use it when answers quote, paraphrase, or act on information retrieved with search_knowledge or get_document.

Rag Deep Dive runs a three-step retrieval workflow: search the corpus, read the top document in full, then find similar documents. Use it for end-to-end or pattern questions that need more than a short snippet.

Rag Troubleshoot searches the knowledge corpus for prior errors, fixes, runbooks, and incidents before general debugging. Use it when a user reports a bug, exception, stack trace, failed CI run, or unexpected behavior.