Agentic Data Kit

Drop one folder. Get a team of AI specialists.

Stop explaining the same things to your AI over and over. The Agentic Data Kit is a curated library of 14 dedicated AI agents, 23 specialized skills, and 6 automated workflows — all pre-configured for Data Engineering, Statistics, Data Science, and Analytical Workflows.

Copy the .agent/ folder into your repository. That's it. Your AI assistant now knows how to build Databricks pipelines, run exploratory analysis, structure experiments, train statistical models, design star schemas, validate data quality, and much more — without you having to teach it anything.

🧠 Why This Exists

In the AI era, your role has shifted. You delegate, orchestrate, and validate — the agents do the heavy lifting. But generic AI assistants don't know your stack, your standards, or your architecture.

This kit solves that. It's pre-built intellectual property for data teams:

• Agents already know their job. A data-engineer agent thinks in Medallion Architecture and Delta Lake. A data-analyst agent thinks in distributions, cohorts, and measurement validity. A data-scientist agent thinks in leakage, calibration, and inference. They don't need to be told how — they already know.
• Skills package the hard knowledge. Instead of pasting documentation into a chat window, skills like databricks-patterns, python-data, and data-quality-testing give the AI deep, structured expertise on specific tools and methods.
• Self-validating work. The kit goes beyond "generate and hope." Built-in validation scripts automatically check architecture compliance, schema integrity, data quality, and idempotency — so the AI also handles the validation step that used to be yours.

🎯 What's Inside

Agents — Your Dedicated Specialists

| Agent | What It Does | |---|---| | data-engineer | Builds ETL pipelines, Databricks notebooks, Delta Lake tables | | analytics-engineer | Designs star schemas, writes dbt models, creates semantic layers | | powerbi-developer | Edits Power BI reports, writes DAX, builds TMDL models | | business-analyst | Frames hypotheses, defines estimands, structures analysis briefs | | data-scientist | Builds ML pipelines, experiments, forecasts, and causal analyses | | data-governance | Enforces data quality, contracts, and compliance | | database-architect | Designs database schemas, indexing strategies, and data models | | data-analyst | Explores data, profiles distributions, investigates anomalies, and builds analytical readouts | | orchestrator | Coordinates multiple agents for complex, multi-domain tasks | | project-planner | Breaks down projects into phases with structured plans | | debugger | Systematic root-cause analysis with evidence-based fixes | | documentation-writer | Produces data dictionaries, runbooks, and technical docs | | explorer-agent | Navigates and maps unfamiliar codebases | | agent-creator | Builds new agents and skills for the kit itself |

Skills — Deep, Pre-Packaged Expertise

| Skill | Domain | |---|---| | databricks-patterns | PySpark, Delta Lake, Unity Catalog best practices | | tmdl-modeling | Tabular Model Definition Language for Power BI | | pbip-report-hacking | Programmatic editing of .pbip report files | | powerbi-semantic-mcp | Power BI semantic model via MCP & REST API | | database-design | Star schemas, indexing, normalization patterns | | clean-code | Idempotency, WAP pattern, testing pyramid | | data-quality-testing | Data contracts, Great Expectations, dbt tests | | architecture | Medallion Architecture, Data Mesh, dimensional modeling | | python-data | Pandas, Polars, NumPy engineering patterns | | plan-writing | Structured task planning with dependencies | | ... and 13 more | Brainstorming, debugging, deployment, code review, etc. |

Workflows — One-Command Automation

| Command | What Happens | |---|---| | /plan | AI creates a phased project plan before writing any code | | /debug | Systematic 4-phase debugging with root cause analysis | | /brainstorm | Compare analytical and modeling approaches before committing | | /orchestrate | Multi-agent coordination for complex tasks | | /test | Generates and runs tests following the testing pyramid | | /status | Shows project progress and task tracking |

🚀 Quick Start

1. Run the Installer

In your terminal, navigate to your project folder and run the interactive CLI. It will ask you which version of the kit you want to install:

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

Press Enter, choose your AI assistant (Antigravity, Copilot, or Claude), and the correct folder will be dropped into your project.

2. Work

Start asking your AI to build things. The agents are already loaded and ready.

You: "Build an ETL pipeline for customer data using Medallion Architecture"
AI:  🤖 Applying @data-engineer... [builds complete Bronze → Silver → Gold pipeline]

You: "Create a star schema for the customer behavior domain"
AI:  🤖 Applying @analytics-engineer... [designs fact and dimension tables]

You: "Investigate churn with cohorts and then propose a prediction strategy"
AI:  🤖 Applying @data-analyst + @data-scientist... [profiles the data, frames hypotheses, proposes the model path]

🔄 Multi-Platform Support

The kit natively works with Antigravity (autonomous agents). If you also use GitHub Copilot or Claude Code, a sync script compiles the .agent/ source into their formats:

python scripts/sync_agents.py

This generates .github/ and .claude/ folders automatically. See CONTRIBUTING.md for details.

📦 Publish Your Own Installer

This fork is already prepared to be published as your own npm package.

npm install
npm run sync
npm run pack:check
npm publish

After publishing, install it in other repositories with:

npx @zlanchus/agentic-data-kit@latest init --target copilot

Detailed publishing notes are in PUBLISHING.md.

🤝 Contributing

Want to add a new agent or skill? Read the CONTRIBUTING.md guide. The golden rule: never edit .github/ or .claude/ directly — always work inside .agent/.

🙏 Acknowledgments

This project was heavily inspired by the pioneering work of vudovn/antigravity-kit in the software engineering space. We've adapted and expanded upon those foundational concepts to create a dedicated solution for Data Engineering and Analytics teams.

📄 License

MIT © Lucca Pinto