OSS · APACHE-2.0 · v1.0.54

Teach your AI coding agent Kinetica.

Knowledge skills that give any agent Kinetica's SQL dialect, Python SDK, and interactive CLI — then activate the right one automatically based on what you're doing. One install, works across Claude Code, Cursor, Codex, and more.

The skills

Two skills. The agent picks.

Both install together. Each skill's frontmatter tells the agent when to activate — so a Python question loads the SDK skill, while a SQL, graph, or admin question loads the execute skill. You don't choose; the agent does.

kinetica-code💻 App developers

Build applications on Kinetica

Teaches the Python SDK (gpudb), the REST API, data pipelines, and embedded SQL — the surface you write application code against.

10 reference docs · activates on SDK + app-development questions
kinetica-execute⚡ All users

Query, analyze, administer

Covers SQL analytics, graph, geospatial, time-series, visualization, and security & admin — with a live dual-runtime CLI that connects, runs queries, and returns results in chat.

26 reference docs · activates on SQL, analytics & interactive ops
Install

One command auto-detects your agent.

The universal installer figures out which agent you're running and drops the skills in the right directory. Claude Code users can use the plugin marketplace; any agent can copy the directories manually.

universal · recommended
npx
# auto-detects your agent
npx skills add kineticadb/agent-skills
Claude Code · marketplace
plugin
/plugin marketplace add kineticadb/agent-skills
/plugin install kineticadb@kinetica-skills
manual · any agent
cp
# pick the directory convention your agent uses
cp -r skills/kinetica-execute .claude/skills/    # Claude Code
cp -r skills/kinetica-execute .cursor/skills/    # Cursor
cp -r skills/kinetica-execute .agents/skills/    # Codex, Windsurf, Roo, etc.
Works with

Any agent that reads a skills directory.

The skills follow the standard SKILL.md directory convention, so they're portable across the major coding agents — and any future one that reads skills from disk.

Claude CodeCursorOpenAI CodexWindsurfGemini CLIGitHub CopilotRoo CodeClineAiderContinueAmazon Q+ any SKILL.md reader
How it works

Small in context. Deep on demand.

A two-tier loading strategy keeps the agent fast on simple questions and thorough on complex ones. The entry point is always loaded; the detailed references load only when a question actually needs them.

Tier 1 · always loaded

SKILL.mdsmall

The activation trigger, critical rules, and links to reference files — small enough to stay in the agent's context without eating the token budget.

Tier 2 · on demand

references/lazy

Detailed domain docs the agent reads only when needed. A simple GROUP BY won't pull geospatial refs; an ST_CONTAINS question will.

knowledge pipeline
build.sh
# 26 source files (~7,400 lines) in one knowledge/ tree
knowledge/  →  build.sh reads each skill's REFS manifest
           →  distributes refs into skills/*/references/
In practice

Just ask. The right skill wakes up.

After installing, prompt your agent naturally. The frontmatter triggers route each request to the skill that fits — no flags, no mode-switching.

activates kinetica-code

Write a Python script that bulk-inserts sensor data using the gpudb SDK.

activates kinetica-execute

Find all delivery trucks within 5 km of the warehouse in the last hour.

activates kinetica-execute

Show me the EXPLAIN plan for this query and suggest index improvements.

activates kinetica-execute

Run SELECT COUNT(*) FROM vehicle_tracks against my Kinetica instance.

The execute skill ships a live CLIA dual-runtime (Node.js + Python) CLI connects to your database, runs queries, generates visualizations, and returns results in the chat.

Give your agent Kinetica fluency.

npx skills add kineticadb/agent-skills

Frequently asked questions

Which coding agents do the skills work with?
Any agent that reads a SKILL.md directory — Claude Code, Cursor, OpenAI Codex, Windsurf, Gemini CLI, GitHub Copilot, Roo Code, Cline, Aider, Continue, Amazon Q, and others. The universal installer auto-detects which agent you're running and drops the skills in the right place.
Do I have to pick which skill to use?
No — the agent does. Both skills install together, and each one's frontmatter tells the agent when to activate: a Python SDK question loads kinetica-code, while a SQL, graph, geospatial, or admin question loads kinetica-execute. There are no flags or mode-switching; you just ask naturally.
What's the difference between kinetica-code and kinetica-execute?
kinetica-code is for writing application code — the gpudb Python SDK, the REST API, data pipelines, and embedded SQL (10 reference docs). kinetica-execute is for querying, analyzing, and administering — SQL analytics, graph, geospatial, time-series, visualization, and security & admin (26 reference docs) — and it ships a live CLI.
How do the skills stay small in the agent's context?
A two-tier loading strategy. The small SKILL.md (the activation trigger, critical rules, and links to reference files) is always loaded, while the detailed references/ docs load only on demand. A simple GROUP BY won't pull geospatial refs; an ST_CONTAINS question will.
Can the execute skill actually run queries against my database?
Yes. The execute skill ships a dual-runtime (Node.js + Python) CLI that connects to your Kinetica instance, runs queries, generates visualizations, and returns the results directly in the chat.
How do I install them, and are they open source?
They're open source under Apache-2.0 (on GitHub). Run npx skills add kineticadb/agent-skills to auto-detect your agent; Claude Code users can use the plugin marketplace, and any agent can copy the skill directory manually into .claude/skills/, .cursor/skills/, or .agents/skills/.

To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions. Cookie Policy