Menú
Provide traceability, source attribution, and explainable results for RAG systems.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Use this skill when building, debugging, or improving Retrieval-Augmented Generation systems, including chunking, vector database selection, hybrid search, reranking, multimodal RAG, code documentation RAG, retrieval latency, and production RAG architecture.
Chunk nested documents into parent-child levels so retrieval can move from broad sections to fine-grained passages.
Use semantic boundaries and embedding similarity to chunk text for higher-relevance retrieval.
Route RAG chunking decisions across semantic, hierarchical, sliding-window, contextual-header, and framework-selection strategies.
Use overlapping windows to preserve context across chunk boundaries while controlling retrieval size.
Reduce retrieval latency with caching, batching, and index-level optimization.
| name | explainable-retrieval |
| title | Explainable Retrieval with Citations |
| description | Provide traceability, source attribution, and explainable results for RAG systems. |
| allowed-tools | ["Read","Grep","Glob","Bash"] |
| category | retrieval-strategies |
| tags | ["explainability","citations","traceability","source-attribution"] |
Explainable retrieval adds citations, source attribution, and traceability to RAG systems. Users can see which documents support each claim, verify information sources, and understand how responses were generated - critical for applications requiring trust and accountability.
Traditional RAG systems lack transparency:
Maintain links from retrieved content to source documents throughout the pipeline.
Why: Without source tracking, it's impossible to provide accurate citations. Every piece of retrieved content must maintain its provenance.
Store source identifiers, titles, URLs, and positions as metadata for each retrieved document.
Generate responses with explicit support for each claim.
Why: Claim-supported generation produces responses that are structured for citation, with each claim clearly linked to supporting evidence.
Direct the LLM to generate responses with inline citation markers (e.g., [Source 1]) for each claim.
Extract citation information from generated responses.
Why: Parsed citations enable structured output that can be displayed as clickable links, verified against sources, and used for filtering.
Parse the generated response to extract claims and their associated source identifiers.
Assess how well each claim is supported by its cited sources.
Why: Confidence scoring helps users understand which claims are well-supported and which might be weakly supported or potentially hallucinated.
For each claim-source pair, have the LLM rate the strength of support on a numeric scale.
Combine claims, citations, and confidence scores into a structured response.
Why: A structured response format enables clean display of information with clear attribution, making it easy for users to verify and understand.
Create a structured response containing summary, individual claims with citations, confidence scores, and source information.
Format citations for user-facing display.
Why: Clear citation display makes it easy for users to click through to sources, verify information, and understand response provenance.
Format the structured response with clickable links, visual indicators of confidence, and clear association between claims and sources.
Combine all components into a cohesive system.
Why: A complete pipeline ensures that every aspect of the response - from claims to citations to confidence scores - is properly tracked and displayed.
For implementation patterns, see LlamaIndex node postprocessors, the Attributed Quote Generation paper, and the LangChain RAG guide.
