Skip to content
Beta Multi-model support (OpenAI, Gemini, and more) launching soon.
RBA RBA Skills

kb-assistant-builder

v0.1.0 ExperimentalNo security review by Dakota Smith

Install

Pick the surface you use. Plugins run inside Claude Code, which is available in the terminal CLI, the desktop app on Mac and Windows, the web app at claude.ai/code, and the VS Code and JetBrains extensions. The slash commands are identical on every surface.

From inside any Claude Code session — terminal, desktop app, web app, or IDE extension:

  1. Add the RBA Skills marketplace (only needed once per machine):

    /plugin marketplace add rbaconsulting/rba-skills-plugin

  2. Install kb-assistant-builder:

    /plugin install kb-assistant-builder@rba-ai-hub

  3. Activate it without restarting:

    /reload-plugins

Overview

KB Assistant Builder

RBA Consulting's end-to-end toolkit for designing, evaluating, and deploying retrieval-augmented generation (RAG) knowledge base assistants — for internal team use or as client-facing deliverables. Built on Azure AI Search + Azure OpenAI as the canonical retrieval backend, with swappable content source adapters for Sitecore XM Cloud, Optimizely CMS, Umbraco, SharePoint, and standalone document libraries.

What's inside

Artifact Purpose
kb-source-ingestion skill Audits raw content sources (SharePoint, CMS exports, blob storage, PDFs), defines a chunking and metadata enrichment strategy, and produces an ingestion pipeline spec for Azure AI Foundry or custom indexers.
kb-rag-pipeline-architect skill Designs the full RAG architecture: embedding model selection, index schema, hybrid/vector retrieval strategy, re-ranking, prompt template, and security trimming — from first spike to production.
kb-qa-evaluator skill Runs a structured evaluation harness against a deployed or local RAG endpoint: faithfulness (groundedness), answer relevance, context precision/recall, and hallucination detection — with a pass/fail gate report.
kb-assistant-packager skill Packages the tested assistant for deployment: generates Bicep/ARM for Azure infrastructure, Teams bot manifest, OpenAPI tool definition (for Claude/Copilot Studio plugin), and an operator runbook.
kb-deployment-runbook skill Produces a day-2 operations runbook covering index re-ingestion cadence, drift monitoring, cost dashboards, incident escalation, and rollback procedures.
kb-quality-reviewer agent Reviewer subagent: audits a RAG pipeline design or evaluation report against grounding, security, and cost standards before client handover.
/build-kb-assistant command Guided entry point — collects source type, audience, platform context, and deployment target, then routes to the right skill sequence.

When to use

Use this plugin when:

  • A client engagement needs a searchable Q&A assistant grounded in their own content (policies, product docs, support articles, project knowledge).
  • An internal RBA team needs to build a delivery-side knowledge base (SOW templates, methodology docs, project retrospectives) with an AI assistant front-end.
  • You are standing up a new Azure AI Search index from a Sitecore, Optimizely, or Umbraco content export and need a retrieval-tested evaluation before go-live.
  • A managed-services engagement requires a repeatable index re-ingestion process and a cost/quality monitoring runbook.

Content source adapters

All skills produce platform-agnostic pipeline specs with named adapter sections for:

  • SharePoint / Microsoft 365 — Graph API connector, item-level permission trimming, incremental crawl via change token
  • Azure Blob Storage — Document Intelligence OCR for PDF/DOCX, blob lifecycle policy for stale content
  • Sitecore XM Cloud — Content API / GraphQL export, field-level metadata mapping to index schema, media asset exclusion rules
  • Optimizely CMS — Content Delivery API JSON export, segment metadata preservation, multi-language handling
  • Umbraco — Content Delivery API v2 export, Examine index bypass, custom document type field mapping

How to install

claude plugin install kb-assistant-builder@rba-ai-hub

After installation, start with the guided entry point:

/build-kb-assistant

Or invoke individual skills directly:

/kb-source-ingestion "SharePoint intranet + Sitecore XM Cloud product docs"
/kb-rag-pipeline-architect
/kb-qa-evaluator

Azure services used

Service Role
Azure AI Search Vector + hybrid retrieval, semantic ranking, security filter trimming
Azure OpenAI Embedding generation (text-embedding-3-large) + answer generation (GPT-4o)
Azure AI Foundry Document Intelligence (OCR), integrated vectorization, agentic retrieval
Azure Blob Storage Raw document staging and incremental change detection
Azure Bot Service Teams and Web Chat deployment channel
Microsoft Entra ID Row-level security, user identity pass-through to search filters

Security and compliance

All generated pipeline specs apply document-level security trimming by default (Entra ID groups → Azure AI Search security filters). No content leaves the tenant boundary. Skills never reference or request credential files.

Bundled artifacts

Skills

  • kb-assistant-packager — Packages a tested RAG knowledge base assistant for deployment: generates production-ready Bicep infrastructure as code, an Azure Bot Service + Teams manifest for channel publishing, an OpenAPI tool definition for Copilot Studio or Claude plugin integration, a client-handover summary, and an operator quickstart. Use after the assistant has passed the kb-qa-evaluator gate — "package our HR policy assistant for Teams", "generate the Bicep IaC for our client KB deployment", or "produce a Copilot Studio plugin definition for our product knowledge assistant".
  • kb-qa-evaluator — Runs a structured evaluation harness against a deployed or local RAG knowledge base assistant: generates a golden question set, scores answers for faithfulness (groundedness), answer relevance, context precision, and hallucination presence, then produces a pass/fail gate report with actionable remediation. Use before promoting a RAG pipeline to production, after index re-ingestion, or as part of a scheduled quality review — "evaluate our KB assistant before launch", "run a faithfulness check after we updated the SharePoint index", or "generate a golden Q&A set for our policy assistant".
  • kb-rag-pipeline-architect — Designs the full retrieval-augmented generation (RAG) pipeline for a knowledge base assistant: retrieval strategy (vector, hybrid, agentic), re-ranking, prompt template, answer grounding rules, and security architecture. Produces an Azure AI Search + Azure OpenAI reference architecture with Bicep infrastructure stubs and a system prompt template ready for evaluation. Use after the ingestion spec is complete — "design the RAG pipeline for our ingested SharePoint KB", "what retrieval strategy suits our multi-source assistant?", or "write the system prompt for our internal policy assistant".
  • kb-source-ingestion — Audits raw content sources for a knowledge base assistant — SharePoint sites, Sitecore/Optimizely/Umbraco CMS exports, Azure Blob Storage document libraries, or file system dumps — and produces a complete ingestion pipeline specification: source connector config, chunking strategy, metadata enrichment schema, and index field mapping for Azure AI Search. Use when starting a RAG assistant build: "plan the ingestion pipeline for our SharePoint intranet", "how do I chunk our Sitecore product docs?", or "design the indexer for this PDF library".
  • kb-deployment-runbook — Produces a day-2 operations runbook for a live knowledge base assistant: index re-ingestion cadence and procedures, content drift monitoring, cost and token usage dashboards, performance SLA tracking, incident escalation paths, and rollback procedures. Use when handing over a deployed assistant to a client ops team, setting up managed services support, or formalising maintenance responsibilities — "write the runbook for our client's SharePoint KB assistant", "create the ops guide for our deployed RAG pipeline", or "document the re-ingestion procedure".

Agents

  • kb-quality-reviewer — Reviews a RAG pipeline design document, evaluation report, or ingestion specification against RBA's grounding, security, and cost standards before client handover. Returns a structured critique with verdict, gap list, and concrete fixes. Use as a review gate on any kb-assistant-builder deliverable before it goes to a client or into production.

Slash Commands

  • /build-kb-assistant — Guided entry point for the KB Assistant Builder plugin. Collects source type, assistant audience, platform context, and deployment target, then routes to the right skill sequence: kb-source-ingestion → kb-rag-pipeline-architect → kb-qa-evaluator → kb-assistant-packager → kb-deployment-runbook. Use to start any new knowledge base assistant build or to identify which individual skill to run for a specific task.
Technical details
Homepage
https://rba-skills.tylerschroeder.net/plugins/kb-assistant-builder/
License
MIT
Maturity
experimental
Security review
No security review
Owners
  • delivery
Platforms
azure, microsoft-365
Disciplines
engineering, delivery, strategy
Lifecycle
plan, build, test-and-review, operate
Tags
  • knowledge-base
  • rag
  • assistant
  • azure-ai-search
  • azure-openai
  • embeddings
  • chunking
  • retrieval
  • chatbot
  • sharepoint
  • teams-bot
  • sitecore
  • optimizely
  • umbraco
  • evaluation
  • faithfulness
  • grounding
  • document-ingestion
  • vector-search
  • hybrid-search
Last updated
by Tyler Schroeder
Source
plugins/kb-assistant-builder/

Edit this plugin's README on GitHub — opens GitHub's web editor. No git knowledge needed.