Welcome to Claude Kit 🚀
This repository is the central library for all AI prompts and agent configurations used across the organization. Its purpose is to foster collaboration, maintain high standards, and accelerate our work by sharing effective and well-tested prompts and specialized agents.
Installation
Install skills using skills tool: npx skills add keboola/ai-kit
Alternatively, install via the Claude Code plugin marketplace:
/plugin marketplace add keboola/claude-kit
After installation, enable the plugins you need:
/plugin install component-developer
Repository Structure
The repository is organized into a plugin-based architecture to make prompts and agents easy to discover and use:
claude-kit/
├── .claude-plugin/
│ └── marketplace.json # Marketplace configuration
├── plugins/
│ ├── component-developer/ # Keboola Python component development
│ ├── dataapp-developer/ # Data app development & deployment for Keboola
│ ├── keboola-cli/ # Keboola project management and review
│ ├── keboola-git/ # Keboola-managed Git (Forgejo) access for data apps
│ ├── powerbi-to-sl/ # Migrate Power BI semantic models to Keboola
│ └── sl-toolkit/ # Semantic layer inspect, validate, build + conversational CRUD
├── README.md # This file
└── LICENSE # MIT license
Not every published plugin lives under plugins/. kbagent is an external
entry: it is published by this marketplace but sourced from
keboola/cli (plugins/kbagent), pinned to a
release tag. It installs like any other plugin here — there is no second
marketplace to add.
Available Plugins
Component Developer Plugin
Location: ./plugins/component-developer
A specialized toolkit for building production-ready Keboola Python components following best practices and architectural patterns.
Features:
- 🎯 Skills: Build component, build UI, debug, test, VCR test, review, backward compatibility review, migrate to UV, getting started
- ⚡ Commands: Init, run, fix, review, migrate-repo, schema-test
- 🔌 MCP Server: Keboola integration
- 📋 Configuration Schemas: JSON Schema with UI elements
- 🚀 CI/CD Integration: Developer Portal and deployment workflows
→ View Component Developer Plugin Documentation
Data App Developer Plugin
Location: ./plugins/dataapp-developer
A toolkit for building and deploying data apps to Keboola — Streamlit development with validate/build/verify workflow, plus deployment guides for Node.js, Python, and any web framework.
Features:
- 🎯 Skills:
dataapp-development covering both Streamlit and Python/JS apps with 16 topical references, plus semantic-layer-usage for confirming a semantic model's physical columns before querying
- 🚀 App Types: Streamlit (Code or Git), React + Vite + Express (Python/JS default, no CDN), the same with a Python backend
- 🏗️ Prod app + drafts: Python/JS apps build into the existing prod app via a draft, so an app created from the Keboola Apps page is never duplicated; an existing draft is continued, publishing removes the draft and its branch, and a running draft can be checked in the agent's own browser through a preview link
- 💾 Storage: RO workspace (default), RW direct access via Query Service, input mapping (legacy)
- 🔒 Authentication: None / Basic / OIDC / GitHub / GitLab / JumpCloud
- ⚡ Performance: Opinionated DuckDB caching pattern (Python + Node templates included)
- 🎨 Styling: Default Keboola theme + optional bundled React+Vite stack + brand customization paths
- 🤖 Kai Integration: Optional natural-language assistant via
kai-client
- 📦 Templates: 5 runnable starters (Streamlit, Python-only, Node-only, Python+Node, DuckDB cache)
- 🔌 MCP Servers: Keboola (remote HTTP) and Playwright (browser automation)
- 🔌 MCP hosting: host any MCP server (Keboola MCP as the worked example) as a single-tenant data app — bearer + OAuth-shape auth, deploy via kbagent / keboola-git / hosted MCP tools
→ View Data App Developer Plugin Documentation
kbagent Plugin
Location: external — keboola/cli, pinned to a release tag
An AI-friendly interface to Keboola Connection projects via the kbagent CLI — explore configs, jobs and lineage, sync configs as files, manage dev branches, and debug SQL in workspaces.
/plugin install kbagent
Features:
- 🎯 Skills:
kbagent (configs, jobs, lineage, dev branches, workspace SQL, GitOps sync, Storage, data apps, flows, semantic layer), kbagent-cicd-migration (port a kbc pipeline to kbagent), kbagent-promotion-pipeline (dev → prod promotion via GitHub Actions)
- ⚡ Commands:
/keboola, /review
- 🤖 Agents:
keboola-expert, kbagent-pr-reviewer
- 🔗 Sourced from
keboola/cli: the plugin is generated and CI-validated against the live command tree in that repo; its release job opens the PR here that bumps the pin
→ View kbagent Plugin Documentation
Keboola CLI Plugin
Location: ./plugins/keboola-cli
A project management and review toolkit for Keboola projects. Includes CLI sync commands and a 10-agent review team that audits SQL quality, security, performance, financial logic, data quality, and template readiness.
Features:
- 🤖 Agents: 10 specialized review agents (SQL, Config, DWH Architecture, Data Quality, Financial Logic, Semantic Layer, Security, Performance, Template Readiness, Consolidator) + config analyzer
- ⚡ Commands:
/kbc-init, /kbc-pull, /kbc-push, /kbc-diff, /kbc-review
- 🔌 MCP Server: Keboola integration for live project analysis
- 📊 Financial Intelligence: Multi-ERP awareness (NetSuite, SAP, Oracle, D365, QuickBooks, Xero), SaaS metrics, budget variance
- 🔒 Security Audit: Credential scanning, PII detection, GDPR/CCPA compliance checks
- 🦆 DuckDB Transformation: SQL dialect, block orchestration, dynamic backends, Snowflake migration, best practices
→ View Keboola CLI Plugin Documentation
Keboola Git Plugin
Location: ./plugins/keboola-git
Access Keboola-managed Git (Forgejo) repos for python-js data apps via the kbagent CLI — provision repos, mint push credentials, push source with raw git, and deploy, including the ~15MB / HTTP 413 build-at-deploy path.
Features:
- 🎯 Skill:
keboola-git — provision/find the managed repo, mint a one-time git_clone_url, raw clone/push, 413 + build-at-deploy recipe, deploy + verify
- 🏗️ Resolve before you create: an empty
configuration_id makes a second app and a second repo, so the create step is gated on get_data_apps
- ⚡ Command:
/keboola-git-copy to-keboola|to-github — bidirectional GitHub ↔ Keboola git source copy with size guard and scratch-branch safety
- 📦 Build-at-deploy: untrack committed builds (
frontend/.next), build in keboola-config/setup.sh to stay under the 15MB push cap
- 🔒 Credential safety: one-time push secrets kept in shell vars, never committed; no force-push, scratch-branch-only reverse copies
- 🛠️ kbagent-CLI driven: ships no MCP server
→ View Keboola Git Plugin Documentation
sl-toolkit Plugin
Location: ./plugins/sl-toolkit
Semantic layer tools for Keboola — inspect, validate, and build models via the metastore API. Works with or without the Semantic Layer data app.
Features:
- ⚡ Commands:
/sl-show, /sl-validate, /sl-build
- 🗣️ Conversational CRUD: add/edit/remove entities by just asking — the
semantic-layer skill handles it
- 🔑 kbagent-free: show, validate, and all CRUD require only a Storage API token
- 🔄 Cascade rename: editing a metric name auto-updates all constraint references, with rollback on failure
- ✅ Deep validation: phantom field and type-mismatch checks against actual Snowflake schemas
- 🏗️ Greenfield wizard:
sl-build does full schema discovery → SQL analysis → generate → validate → push
- 🌐 Multi-cloud: supports GCP, AWS, and Azure stacks
→ View sl-toolkit Plugin Documentation
MCP Server Setup
Some commands and plugins require MCP (Model Context Protocol) servers to be configured. If MCP tools are not available when running a command, use the /mcp command to authenticate and configure them.
Troubleshooting
If you encounter "MCP tools not available" errors:
- Run
/mcp to see available MCP servers and their status
- Authenticate with the required MCP server
- Re-run your command
Development
Validation
Always validate your changes before committing:
claude plugin validate .
Testing
Test agents and commands locally:
# Test an agent
@agent-name
# Test a command
/command-name
Versioning
We follow semantic versioning. Update version numbers in:
.claude-plugin/marketplace.jsonplugins/<name>/.claude-plugin/plugin.jsonplugins/<name>/README.md
Skill evals
Every PR runs two eval tiers from evals/ (see the
Skill evals workflow):
- Tier 0 — static lint: frontmatter validity, skill
name ↔ directory,
plugin/marketplace version consistency, dangling references.
- Tier 1 — activation: each skill's
description: is measured against
labeled utterances in plugins/<plugin>/evals/<skill>/trigger-evals.json
for routing precision/recall. Every skill must have a case set.
Deeper behavior and live E2E tiers run from
keboola/KaiBench (docs/skill-evals.md).
License
MIT licensed, see LICENSE file.
