Agentic Data Kit

Run a single command. Get a team of AI data specialists.

Lean, opinionated AI agents · skills · workflows for data teams — one source of truth, every assistant.

npx @luccapinto/agentic-data-kit@latest init

5 agents · 8 skills · 3 workflows · 4 platforms

---

Contents

• 🧠 Why this exists
• 🚀 Quick start
• 🎯 What's inside
• 🤖 Agents
• 🧩 Skills
• 📈 Power BI — as code, not clicks
• 🎨 Presentations — as code
• 🔄 How it works (multi-platform)
• 🧭 Design philosophy
• 🤝 Contributing & license

---

🧠 Why this exists

Generic LLMs don't know your stack, your standards, or your architecture — but they already know how to write clean code. So this kit follows one rule:

Capture only the domain-specific context that matters, and nothing the model already does well.

Lean instructions keep the model's attention on what's important — which is exactly what makes agents reliable. The result is a curated library for Data Engineering, Analytics, BI, and data storytelling that drops into your assistant and just works: build pipelines, design star schemas, edit Power BI as code, run quality checks, document models, and turn analyses into on-brand presentations — without re-explaining the basics.

---

🚀 Quick start

npx @luccapinto/agentic-data-kit@latest init

Pick your AI assistant in the prompt and the kit installs itself. Then just ask:

You: "Build an ETL pipeline for customer data using Medallion Architecture"
AI:  Applying @data-engineer → Bronze → Silver → Gold pipeline

You: "Add a revenue measure to the Power BI model"
AI:  Applying @powerbi-developer → writes DAX, edits TMDL, runs the BPA

You: "Turn this analysis into a McKinsey-style PDF"
AI:  Applying @presentation-designer → builds the report, exports to PDF

🌍 Instructions are written in English for portability; the assistant replies in your language automatically.

---

🎯 What's inside

	Count	What it is
🤖 Agents	5	Role-based specialists with isolated context and clear domains
🧩 Skills	8	On-demand knowledge modules that self-activate by description
🔄 Workflows	3	Slash-command procedures (/plan, /validate-pbi, /document-dashboard)
🖥️ Platforms	4	Claude Code · GitHub Copilot · OpenCode · Cursor (from one source)

---

🤖 Agents

Focused specialists — each owns a distinct domain, so routing is never ambiguous.

Agent	What it does
data-engineer	ETL/ELT pipelines, Medallion architecture, idempotency, Write-Audit-Publish
analytics-engineer	Star schemas, dbt models, warehouse architecture & performance
data-scientist	Analysis, metrics, dashboards & semantic review — through to ML, statistics, forecasting, A/B testing
powerbi-developer	Power BI as code — TMDL semantic model and PBIR reports, validated with Tabular Editor 2
presentation-designer	Analyses → presentations as code — decks, interactive sites, consulting PDFs, editable PPTX — on-brand via DESIGN.md

🛡️ Governance is cross-cutting, not a role. PII masking, Write-Audit-Publish, contracts, and downstream-impact checks live in the always-on workspace rules and apply to every agent.

---

🧩 Skills

Skills activate themselves from their description (progressive disclosure) — no agent required, no need to name them. Ask "document this" and the docs skill fires on its own.

Skill	Use it for
pbi-semantic-layer-tmdl	Author Power BI semantic models in TMDL
pbi-report-layer-pbir	Edit Power BI reports as code (pages, visuals, themes)
pbi-quality-rules	Real Best Practice Analyzer via the free Tabular Editor 2
building-html-presentations	Decks, scrollable decks, interactive sites & consulting PDFs + an 18-chart cookbook
generating-pptx	Native, editable PowerPoint via python-pptx (real charts, not images)
applying-visual-identity	Apply a company brand from a DESIGN.md (Google Labs spec)
documentation-templates	Runbooks, data dictionaries, metric defs, ADRs, Power BI catalog
creating-agents-and-skills	Governance gate — add an agent/skill only when it helps, and keep every tool folder in sync

---

📈 Power BI — as code, not clicks

The powerbi-developer agent edits PBIP projects directly on disk:

• 🧱 TMDL for the semantic model (tables, measures, relationships, RLS)
• 📊 PBIR for reports (pages, visuals, themes, bookmarks)
• ✅ Validated with the free Tabular Editor 2 Best Practice Analyzer — no paid tooling, no Fabric required

Microsoft's official powerbi-modeling-mcp (preview) is supported as an optional path for bulk refactors, but file editing is the default.

---

🎨 Presentations — as code

The presentation-designer agent turns analyses, EDAs, and business plans into polished, on-brand deliverables — and reads your brand from a DESIGN.md so everything stays consistent.

Starter	Output
reveal-deck.html	Dashboard-style slide deck (reveal.js) → HTML + PDF
flex-deck.html	Flexible scrollable deck with dynamic visual toggles
interactive-site.html	Self-serve scrolling report site
report-pdf.html	McKinsey/Deloitte-style PDF (action titles, exhibits, callouts)
marp-deck.md	Quick Markdown → PDF/PPTX
charts-reference.html	Chart cookbook — 18 ready-to-copy chart patterns

Plus generating-pptx for native, editable PowerPoint (real text & charts, not images) via python-pptx.

---

🔄 How it works (multi-platform)

.agent/ is the single source of truth. One sync script compiles it for every assistant in that tool's native format — write once, run everywhere.

python scripts/sync_agents.py

Target	Output
Antigravity	.agent/ (native source)
Claude Code	.claude/ (agents · skills · workflows · commands/) + CLAUDE.md
GitHub Copilot	.github/ + copilot-instructions.md
OpenCode	.opencode/ (agents mode: subagent · skills · commands) + root AGENTS.md
Cursor	.cursor/ (rules/*.mdc · commands/) + root AGENTS.md
OpenAI Codex / cross-tool	root AGENTS.md

.agent/                ← edit here (source of truth)
├── agents/            # 5 specialist personas
├── skills/            # 8 on-demand knowledge modules
├── workflows/         # 3 slash commands
└── rules/             # global rules (rules.md)

.claude/  .github/  .opencode/  .cursor/  AGENTS.md   ← auto-generated (do not edit)

---

🧭 Design philosophy

• Less is more. Extra agents and verbose instructions cost the model attention and blur routing. We capture domain-specific context only — never restate what the model already knows.
• Progressive disclosure. Skills load when relevant, not upfront.
• No ambiguous routing. Each agent owns a distinct domain.
• Idempotent & safe by default. Write-Audit-Publish, mask PII at Silver, check downstream impact.
• Self-orchestrating. Agents route, skills self-activate, governance is always on.

---

🤝 Contributing & license

Want to extend it? See CONTRIBUTING.md — especially the "should this agent exist?" gates. The creating-agents-and-skills skill walks you through deciding, authoring, and syncing a new capability across every platform.

MIT © Lucca Pinto

_{Built for data teams who'd rather ship than re-explain the basics.}