How this whole thing works, under the hood.

The four systems

1What's actually running

Four systems talk to each other during the workshop. You only ever talk to the first one — Claude Code. It talks to the other three on your behalf.

Sentence → dashboard

2The workflow, step by step

You type "build me a wholesale sales command dashboard." Six things happen in the next 45 seconds.

What's a "skill"?

3Skills — Claude's domain knowledge modules

A skill is a markdown file that lives in .claude/skills/<name>/SKILL.md. Each has a header of "trigger phrases" — natural language patterns like "build me a dashboard for retail regional ops". When Claude Code sees a match in your prompt, it auto-loads that skill's full body: domain knowledge, references, checklists, code patterns, and known-gotcha workarounds. The skill is what turns a general-purpose LLM into a subject-matter expert on Tableau Next authoring, semantic modeling, or Agentforce wiring.

You never invoke a skill by name. You just describe what you want. The router does the rest.

The 22 skills in this repo

What's an LWC?

4Lightning Web Components — the widget primitives

A Lightning Web Component (LWC) is Salesforce's native web-component framework — think custom HTML elements that render inside any Salesforce surface, including a Tableau Next dashboard. On this dashboard stack, LWCs are the widgets: a KPI card, a treemap, a radial gauge, a map, a sparkline. Each is a bundle of four files:

force-app/main/default/lwc/heroKpiCard/
  ├── heroKpiCard.js        ← the logic (SDK queries, filter subscription)
  ├── heroKpiCard.html      ← the template
  ├── heroKpiCard.css       ← the styles
  └── heroKpiCard.js-meta.xml  ← where it can be placed (dashboard, record page, app)

The repo ships 142 canonical LWC bundles — bar charts, heatmaps, treemaps, waterfall, funnel, KPI cards, sparklines, choropleth maps (Leaflet raster-only, since Mapbox is blocked by Lightning Web Security), sentiment cards, and Agentforce-aware insight cards. When Claude picks an LWC for your dashboard, it references it by bundle name in the payload's customComponent block. You can browse them at /lwcs/.

Why is there Python?

5Python — what Claude runs on your behalf

You never write Python. But when Claude Code says "deploying the dashboard now," what's actually running is a Python script in bootstrap/. Python is the glue because Tableau Next's REST API needs a structured payload, and Python's dictionaries map cleanly to JSON. Four scripts do most of the work:

• bootstrap/beat1_dashboard_builder.py — assembles the CloudKicks dashboard JSON
• bootstrap/beat2_dashboard_builder.py — assembles Beat 2 industry dashboards
• bootstrap/lib/dashboard_builder.py — the shared widget/layout/filter library
• bootstrap/teardown.py — nukes an attendee namespace via SOAP destructive-deploy

Each one POSTs to the Tableau Next REST API using auth pulled from sf org display --json. If you want to understand what happened, open the files — they're deliberately readable.

The load-bearing concept

6Semantic Data Model + Data Cloud DLOs

If you learn one thing on this page, learn this one. Tableau Next is a two-layer system.

When Claude "extends the SDM" during Beat 2 by ingesting the CloudKicks_Locations and CloudKicks_CSAT CSVs, it does three things in sequence: (1) creates two new DLOs from the CSVs via the Data Cloud Ingestion API, (2) registers them as Data Model Objects (DMOs) — the SDM's view of a DLO — and (3) adds them to the semantic model so metrics can reference them and joins can be authored. Every step is metadata; nothing is copied.

How filters reach widgets

7Filter propagation — native vs LWC

A dashboard filter chip (Region, Product Line, Time) has to reach every widget on the page. Two mechanisms, one bug-heavy history:

• Native widgets (KPI metric cards, bar charts, tables) subscribe by declaring a receiveFilterSource block in the payload — an array of publishers naming which filter widgets to listen to.
• LWC widgets subscribe through the SDK event bus — the LWC's JavaScript calls sdk.on('filterChange', handler) in connectedCallback, then re-runs its semantic query when the payload arrives.

The AI moment

8Agentforce integration

Every dashboard you build has two Agentforce surfaces:

1. The Concierge box — a natural-language question bar at the top of the dashboard. You type "which reps are trending down this quarter?"; Concierge resolves against the same governed SDM the dashboard reads, so the answer stays inside the metric definitions you authored.
2. The aiDashboardInsight LWC — a narrative insight card that subscribes to filter changes on the dashboard, packages the current filter state + a hero metric into a prompt, calls Agentforce through the AiDashboardBriefingController Apex controller, and renders the returned narrative back into the card. It refreshes automatically when you flip a filter.

Because Agentforce reads the SDM, not the raw DLOs, everything it says respects your row-level security, audience scoping, and metric governance. The audit trail that fires on a click-through action is the same audit trail the dashboard would fire.

What you walk out with

9The take-home repo

When you leave Dallas, you fork github.com/tcufrogger/tableau-vibe-coding-workshop-starter-kit. Everything on your laptop after that is yours to rebrand and re-run for your customers. It contains:

Open the repo on GitHub →

Every acronym, defined

10Glossary

Anchor-linkable. Ctrl-F your way in.

SDM — Semantic Data Model

The governed metadata layer inside Tableau Next / Data Cloud where metrics, calc measurements, calc dimensions, joins, and business-friendly names live. One SDM per workspace. Everything Claude authors during the workshop lands here.

DLO — Data Lake Object

The physical Data Cloud table. Where CSV uploads and CRM ingests land. Row/column storage; no business meaning by itself.

DMO — Data Model Object

The SDM's registered view of a DLO. A DMO gives a DLO business-friendly column names, joins, and semantic role (dimension vs measure). You author dashboards against DMOs, not DLOs directly.

CLC — Calc Measurement (suffix _clc)

A calculated field defined in the SDM. Example: Deal_Health_clc = Amount * Probability / Days_To_Close. Governed, versioned, and referenced everywhere by its api name.

MTC — Metric (suffix _mtc)

A registered metric on the SDM — the promoted form of a calc measurement. Metrics are what Agentforce and Concierge can query semantically ("show me pipeline health by rep"). Every metric declares an aggregation (SUM / AVG / SNAPSHOT).

LWC — Lightning Web Component

Salesforce's native web-component framework. On this stack, the visualization primitives (bar chart, KPI card, map, AI insight card) are LWCs. Repo ships 142 canonical bundles.

CDP — Customer Data Platform / Data Cloud

Same thing, two names. CDP is the generic term; Data Cloud is Salesforce's product name for its CDP. The DLOs live here.

Concierge

The Agentforce natural-language query interface embedded at the top of a Tableau Next dashboard. You type a question; it resolves against the SDM.

Agentforce

Salesforce's agent runtime. Runs the Concierge, powers the aiDashboardInsight card, and executes actions (New Task, Case, Flow) back into the org.

Skill

A markdown file in .claude/skills/<name>/SKILL.md that Claude Code auto-loads when your prompt matches its trigger phrases. Contains domain knowledge, references, checklists, and code patterns.

MCP — Model Context Protocol

Anthropic's open protocol for exposing tools + data sources to an LLM. Claude Code speaks MCP; some skills invoke MCP servers (Tableau Desktop, codesearch, GitHub) to reach out to external systems.

Payload

The JSON POST body sent to the Tableau Next REST API to create a dashboard, metric, or LWC binding. Assembled by the Python dashboard-builder library.

Hot Pass

Salesforce's internal provisioning tool used to hand each workshop attendee their pre-configured OrgFarm scratch org + AWS Workspace at check-in. Attendees see a code (e.g., TCRFPEDN); they never touch Hot Pass directly.

OrgFarm

Salesforce's internal template-org system. Provisions a scratch org with a pre-baked image (Tableau Next enabled, SDM shipped, LWCs deployed, sample data loaded). Your workshop org is an OrgFarm pull.

Beat 1 / Beat 2

The workshop's two teaching phases. Beat 1 (40 min) is the shared CloudKicks story — everyone in the room builds the same dashboard together. Beat 2 (50 min) is "make this yours" — you pick one of the 25 industry data kits and Claude builds a dashboard for your industry.

Vibe coding

Talking to Claude Code in plain English to build software. No syntax. No template. Describe intent; the AI produces working code. The workshop's core practice.

Filter propagation

How a filter chip's value (Region = West, Time = Last Quarter) reaches every widget on a dashboard. Native widgets use receiveFilterSource; LWCs use the SDK event bus. Both are wired automatically by tableau-next-author.

Cross-viz filter

When clicking one chart (say, a stage in a bar chart) filters every other chart on the page. Different subscription mechanism from the top-of-dashboard filter chips, though the payload shape overlaps.

Data-viz-first / semantic-first

The two authoring modes in Tableau Next. Semantic-first (what this workshop uses) starts from a governed metric on the SDM and generates the viz. Data-viz-first starts from a chart and reaches down to raw data — less governed, faster to prototype.

LWS — Lightning Web Security

The runtime sandbox that Salesforce puts around every LWC. Blocks a lot of common web APIs (Blob-URL Web Workers, some fetch shapes) — which is why Mapbox GL / MapLibre GL don't work inside a Tableau Next LWC and Leaflet-with-raster-tiles is the only interactive-map path.

sf CLI

Salesforce's official command-line tool. Handles the OAuth login (sf org login web), returns access tokens, and is how every script in this repo talks to the org. No Connected App required.

Wave

The internal shorthand for a build/release cycle on this repo (e.g., "Wave F2b shipped the AI briefing controller"). You'll see it in commit messages and doc filenames.