Spice Skills for AI Agents -- Spice AI Blog

We're excited to introduce Spice Skills, an open-source collection of packaged instructions for AI coding agents working with Spice.ai OSS. Each skill covers a specific capability -- federated SQL queries, data acceleration, vector and full-text search, LLM inference, and MCP tool use -- so agents can configure and operate Spice correctly without manual prompting.

Skills follow the open Agent Skills format and work with Claude Code via Claude Skills, Cursor, and any agent that supports the standard.

What's Included

The initial release covers the core Spice workflows:

Skill	What the agent can do
`spice-setup`	Install Spice, initialize a project, and run the runtime
`spice-connect-data`	Connect to data sources and query across them with federated SQL
`spice-acceleration`	Accelerate data locally for sub-second query performance
`spice-search`	Search with vector similarity, full-text keywords, or hybrid RRF
`spice-ai`	Add AI capabilities -- chat, text-to-SQL, tools, memory, model routing
`spice-caching`	Cache query and search results with TTL and stale-while-revalidate
`spice-secrets`	Manage credentials with secret stores
`spice-text-to-sql`	Convert natural language to SQL using the /v1/nsql endpoint
`spice-terraform`	Provision and manage Spice infrastructure with Terraform

Each skill ships as a SKILL.md -- instructions the agent reads directly -- along with optional helper scripts and example configurations.

Demo of basic usage with Spice Skills

Installation

Claude Code (Claude Skills):

/plugin marketplace add spiceai/skills
/plugin install spice@spiceai-skills

Skills become available as /spice:spice-setup, /spice:spice-ai, /spice:spicepod-config, and so on.

Other agents (npx):

npx skills add spiceai/skills

Once installed, skills are invoked automatically when the agent detects a relevant task -- no manual prompting required.

Why Skills

Working with a new runtime in an AI coding agent typically means copy-pasting docs, correcting hallucinated API shapes, and re-explaining configuration patterns. Skills solve this by giving the agent accurate, structured knowledge up front.

For Spice specifically, this matters because the configuration surface is broad: 30+ data connectors, multiple acceleration engines, LLM providers, embedding models, and a growing set of tools. For teams building retrieval-augmented generation (RAG) pipelines or agentic AI applications, a well-scoped skill keeps the agent on the right path without requiring the developer to re-explain Spice's configuration model each session.

Skill Structure

Each skill follows a consistent layout:

spice-setup/
  SKILL.md           # Agent instructions
  scripts/           # Helper automation (optional)
  examples/          # Example spicepod.yaml configurations (optional)

The format is intentionally minimal. SKILL.md is plain text the agent can parse; scripts handle anything that benefits from automation.

Get Started

The skills repository is open source at github.com/spiceai/skills. Contributions for additional connectors, deployment patterns, and Spice Cloud workflows are welcome.

Spice AI Skills FAQ

What is Spice Skills?

Spice Skills is an open-source collection of packaged instructions for AI coding agents working with Spice.ai OSS. Each skill is a SKILL.md file that gives an agent structured knowledge about a specific capability -- so it can configure datasets, connect data sources, run federated queries, and set up models without requiring manual prompting from the developer.

Which AI coding agents support Spice Skills?

Spice Skills work with any agent that supports the Agent Skills format. Claude Code installs via the plugin marketplace (/plugin marketplace add spiceai/skills). Any other agent can add skills using npx skills add spiceai/skills.

Is Spice Skills free to use?

Yes. The repository is MIT-licensed and open source at github.com/spiceai/skills. The AI coding agent tools you use are subject to their own licensing.

How does Spice Skills differ from reading the Spice documentation?

Documentation is written for humans to read. Skills are structured for agents to parse -- scoped to specific tasks, paired with helper scripts and example configs, and automatically invoked when a relevant task is detected. An agent with an installed skill does not need the developer to re-explain Spice's configuration model each session.

Can I build and contribute custom Spice Skills?

Yes. Each skill follows a minimal structure: a SKILL.md with agent instructions, an optional scripts/ directory, and optional examples/. Contributions for additional connectors, deployment patterns, and Spice Cloud workflows are welcome via pull request to github.com/spiceai/skills.

Introducing Spice Skills for AI Agents