Preset Agent Skills: Teaching AI to Think Like a Data Expert

Today, Preset is releasing Preset Agent Skills — an open-source library of installable AI guidance packages that give any AI agent deep, curated knowledge about working with Preset, Apache Superset™, and your data.

Skills are installable instruction sets, not plugins or chatbot templates. They give AI agents domain expertise: the specific workflows, safety boundaries, and tool-routing logic that no general-purpose AI picks up on its own. Whether you're talking to Preset Chatbot, Claude Desktop, Cursor, GitHub Copilot, or another MCP-compatible client, the AI can now operate your analytics environment with the same care and precision as a seasoned data engineer.

Why AI Agents Need Domain Expertise

AI agents are genuinely good at generating SQL, summarizing trends, and explaining charts in plain language. Where they fall short is knowing the terrain.

Without guidance, an AI agent talking to your Preset workspace has no idea that pushing a dashboard to production requires a confirmation step, that guest tokens expire after an hour, or which API endpoint to call for exporting versus overwriting a chart. It will make reasonable guesses — and reasonable guesses in analytics environments can cause real problems.

Preset Agent Skills address this by encoding that operational knowledge directly into the AI's behavior. The agent stops guessing and starts following documented, reviewed, production-tested workflows.

What's in the Skills Library

Preset Agent Skills ships as three installable packages, each covering a different way of working with Preset and Apache Superset.

preset-api-skills covers direct API workflows. Seventeen focused skills spanning authentication, workspace management, dashboard and chart operations, SQL execution, dataset management, embedding, guest tokens, row-level security, roles and permissions, and Snowflake Cortex Agents. When an agent needs to build or modify something in Preset, this package tells it how to do so correctly and in the right order.

preset-mcp-skills covers Preset MCP tool workflows. Eight skills that teach AI agents to use the Superset MCP server for discovery, visualization, SQL execution, and troubleshooting. The package also enforces the right boundaries between MCP and direct API access — a distinction that matters for security and that no AI figures out independently.

preset-cli-skills covers the sup CLI. Two skills that handle read-only workflows separately from mutation workflows, because the difference between exporting dashboards and overwriting them needs to be explicit before the agent starts typing commands.

Each skill is a focused, auditable instruction set. Install the ones relevant to your work.

See how MCP, Chatbot, and skills come together for enterprise-ready AI analytics.

One Skill Package, Every Major AI Tool

The same packages install across Claude Desktop, Claude.ai web, Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Snowflake Cortex Code CLI, and Gemini CLI. Teams using multiple AI tools get consistent behavior across all of them, without maintaining separate configurations for each client.

For organizations already running Preset MCP with Claude, skills upgrade the AI from "can use Superset tools" to "knows how to use them well." For teams using Preset Chatbot, skills form part of the knowledge layer that makes conversational analytics feel grounded rather than generative — the AI is working from documented practice, not inference.

What Changes for Users

The effects are different depending on how you work with Preset — but both audiences benefit.

For analysts and business users, nothing changes on the surface. The everyday user of Preset Chatbot or a connected AI client might not install skills or configure anything. The effects show up in the quality of what the AI does. Answers are faster and more accurate because the agent knows your workspace's authentication model, tool routing, and permission boundaries. A revenue analyst asking "which datasets have columns related to customer churn?" gets a real answer drawn from the actual Superset MCP discovery tools, not a generic query attempt. Self-service becomes more reliable: agents explore and iterate without modifying production resources until explicitly asked to, and the same RBAC and row-level security governing your Preset workspace governs every AI action inside it.

For data engineers and platform teams, skills make AI agents safe to actually use in production workflows. Without them, an agent building a dashboard or running a migration has no concept of operation order, permission scope, or when to stop and confirm. With skills installed, the agent knows which API calls to sequence and how, when to use MCP tools versus direct API access, and where the hard boundaries around destructive operations are. The skill files themselves are plain text — readable, auditable, and forkable. If your organization has its own conventions or restrictions on top of Preset's defaults, you can adapt the skills accordingly and know exactly what instructions the AI is operating under.

The consistency spans across both groups. Whether analysts use Claude, Cursor, or Copilot, and whether engineers are scripting with sup or calling the API directly, the same operational knowledge underlies every interaction. Skills are the common layer.

Open Source, Because the Instructions Should Be Readable

Preset Agent Skills are published under the Apache 2.0 license at github.com/preset-io/agent-skills. Anyone can read them, fork them, contribute to them, or customize them for their own deployments.

AI agents are increasingly taking real actions in production analytics environments — building dashboards, executing SQL, managing workspace memberships. The instructions they follow should be auditable. Organizations should be able to inspect exactly what guidance the AI is operating under, not accept it as opaque behavior. Open source makes that possible.

Getting Started

Preset Chatbot users benefit automatically as workspace updates roll out. No setup required.

Claude Desktop users: Download skill ZIPs from the latest GitHub Release and upload them in Claude Desktop under Settings > Connectors > Skills. Start with preset-mcp.zip and preset-mcp-discovery.zip for most Preset MCP workflows.

Claude Code and OpenAI Codex users: Install from the plugin marketplace:

/plugin marketplace add preset-io/agent-skills
/plugin install preset-api-skills@preset-agent-skills
/plugin install preset-mcp-skills@preset-agent-skills
/plugin install preset-cli-skills@preset-agent-skills

Cursor, GitHub Copilot, Snowflake Cortex Code CLI, and Gemini CLI users: See the full installation guide for client-specific setup — each takes less than five minutes.

Go Deeper

• Preset Agent Skills on GitHub — full documentation, installation guides, and all three skill packages
• Preset MCP: AI That Doesn't Just Answer — It Builds — the MCP foundation these skills build on
• Preset MCP: From Open Source to Enterprise — how Preset extends the open-source Superset MCP service for enterprise use
• Shipping Preset Chatbot: From AI Prototype to Production — how Preset Chatbot was built and what powers it under the hood
• Meet sup: Superset's New CLI for Automation and Agents — the CLI that preset-cli-skills wraps

Join the live event for a walkthrough of how skills give your AI agent real Preset expertise.