Deliver Lyapunov-convergent RAG answers with self-critique and convergence tracking.
npx skills add https://github.com/lofibrainwav/AFO_Kingdom --skill ultimate-rag
Please help me install this Agent Skill. Skill: ultimate-rag Source: https://github.com/lofibrainwav/AFO_Kingdom/tree/main/skills/ultimate-rag Command: npx skills add https://github.com/lofibrainwav/AFO_Kingdom --skill ultimate-rag
This Skill delivers a hybrid Corrective RAG and Self-RAG system with mathematical convergence guarantees, reducing hallucinations and improving retrieval reliability.
rag = UltimateRAG(top_k=5, max_iterations=10) result = rag.query("What is Trinity Score philosophy?") print(result.answer)
High-intent search queries and answers about installing and using this skill.
RAG mitigates hallucination by retrieving factual context from knowledge sources before generation, then applying self-critique loops and relevance thresholds to filter unreliable outputs. Lyapunov convergence guarantees iterative refinement stabilizes toward factually grounded answers within a bounded number of iterations.
Self-RAG applies iterative self-critique to refine retrieved answers by evaluating their own quality and relevance at each step. Convergence tracking ensures the process stabilizes, progressively eliminating weak retrievals and improving factual accuracy through structured feedback loops.
Yes. This Skill includes threshold-based correction that automatically triggers web search fallback when relevance scores fall below confidence limits, and augments results with knowledge graph data to strengthen entity relationships and factual grounding.
Convergence is mathematically bounded by Lyapunov stability guarantees, which limit iterations to a preset maximum—configurable via max_iterations—ensuring predictable completion time while maintaining answer quality improvement across refinement cycles.
The Skill explicitly outputs confidence scores and iteration metrics, exposing convergence progress and answer stability. These metrics let you audit factual reliability and verify that answers have reached a stable, high-confidence state before use.
Vector search is a core functional requirement for relevance scoring. You need access to vector embeddings and similarity search—either via an integrated vector database or external retrieval service—to execute the full retrieval and ranking pipeline.
figuedmundoImplement RAG systems integrating vector databases and semantic search for LLM applications.
bobmatnycInstrument traces, metrics, and logs with OpenTelemetry for distributed systems.
giosuetedeschi-specDesign and optimize Retrieval-Augmented Generation pipelines for improved context retrieval.
aisa-groupOptimize and structure prompts for consistent LLM outputs.
