skills.rest
gabedsam01
gabedsam01COMMUNITY
Explore all skills by this author →
WebsiteGitHub

excalidash-llm-rag-pipeline

Generates RAG pipeline flowchart diagrams with distinct index and query paths via Excalidraw MCP tools.

2|Updated Jun 17, 2026
Search Tags:#llm#rag#mcp#excalidraw#flowchart#diagram#vector store
One-click install
npx skills add https://github.com/gabedsam01/excalidash-v2 --skill excalidash-llm-rag-pipeline-gabedsam01
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: excalidash-llm-rag-pipeline
Source: https://github.com/gabedsam01/excalidash-v2/tree/main/skills/excalidash/excalidash-llm-rag-pipeline
Command: npx skills add https://github.com/gabedsam01/excalidash-v2 --skill excalidash-llm-rag-pipeline-gabedsam01

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Documenting a retrieval-augmented-generation system as a clear diagram is error-prone: the async indexing flow and the live query flow get visually mixed, fallback and guardrail branches are forgotten, and credentials leak into node labels. This Skill produces a validated Excalidraw flowchart of a RAG pipeline with both paths and all failure branches drawn explicitly. ## Core Features & Use Cases - Dual-path flowchart generation: Draws the async index path (ingest → chunk → embed → vector store) with dashed edges and the live query path (retrieve → assemble → prompt → LLM → eval → response) with solid edges in a single canvas. - First-class failure branches: Models the cache-hit short-circuit, retrieval-miss fallback (keyword/BM25 or refuse/clarify), and guardrail/error path as decision nodes that each reach a terminal. - Quality enforcement loop: Runs lint, score, repair, polish, and architecture validation until the drawing scores at least 95 with zero hard blockers, then saves, versions, and exports it. - Secret redaction: Replaces vector-DB URLs, provider keys, and bearer tokens with typed [REDACTED_<TYPE>] placeholders before any tool call and re-scans the export. - Use Case: Ask to diagram a docs-to-pgvector RAG system with a semantic cache and PII guardrail, and receive a shareable, scored SVG/PNG flowchart with both paths visually distinct. ## Quick Start Ask the AI to diagram your RAG pipeline showing the index path separate from the query path, including the cache, retrieval fallback, and guardrail branches.

Frequently Asked Questions about excalidash-llm-rag-pipeline

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I diagram a RAG pipeline with separate index and query paths?▼

Describe your sources, chunking, embedding model, vector store, and query stages, and the Skill creates one flowchart where index-path edges are dashed and query-path edges are solid. A legend keys both styles so readers can distinguish async indexing from live retrieval.

How to show cache, fallback, and guardrail branches in a RAG diagram?▼

The Skill adds decision nodes for cache hit, retrieved context, and guardrail pass. Each branch routes to a terminal: cache hits go straight to the response, retrieval misses go to a keyword/BM25 fallback or refuse/clarify terminal, and guardrail failures go to a safe fallback.

What tools does this Skill need to create Excalidraw diagrams?▼

It requires the Excalidraw MCP server tools, including create_diagram_from_prompt, lint_drawing, score_drawing, repair_drawing, auto_polish_drawing, validate_architecture, and save/export functions. These are declared in the allowed-tools frontmatter.

When should I not use the RAG pipeline diagram skill?▼

Do not use it for MCP server or tool-architecture diagrams, observability telemetry flows, time-ordered sequence diagrams, CI/CD or cloud deployment topologies, or layered clean/hexagonal architectures. Those scenarios have dedicated sibling skills.

Does the RAG diagram skill handle API keys and connection strings safely?▼

Yes. Secrets such as vector-DB URLs, provider keys, and bearer tokens are replaced with typed [REDACTED_<TYPE>] placeholders before any tool call, and the exported drawing is re-scanned as a backstop. Detected secrets are never echoed back.

Why does my RAG diagram score below 95 and how is it fixed?▼

Low scores come from hard blockers like arrow-text intersections or penalties like small fonts and high density near the retriever. The Skill runs a mandatory lint-score-repair loop, rolling back any repair that lowers the score, until it reaches 95 with zero blockers.

Related Skills

Explore More Software Engineering Skills→
langchain-ailangchain-ai
1.1k

LangChain RAG Pipeline

Orchestrate document loading, splitting, embedding, and vector store management for RAG pipelines.

Official
Advanced
466852675466852675

fireworks-tech-graph

Generate validated SVG technical diagrams and export PNG images.

Community
Advanced
rloisellrloisell
7

diagram-generation

Generate and export architectural diagrams in draw.io, PlantUML, and Mermaid formats.

Community
Advanced
johnnystefanjohnnystefan

documentation/uml-interaction-diagrams

Generate Mermaid sequence diagrams documenting object interactions over time.

Community
Intermediate