RevOS Skills

Public, version-controlled home of the RevOS Agent Skills.

Skills are grouped into bundles — one top-level directory per bundle, each a plugin in the .claude-plugin/marketplace.json catalog. Install a whole bundle by pointing your tool at its directory.

Bundles

data-engineering

End-to-end RevOS data-engineering workflow. Installed by default when you scaffold a project with revos init.

Skill	What it does
create-connections	Author a RevOS Connection YAML to sync a Source into the warehouse.
load-sample-data	Populate a BigQuery dataset with sample data via bq cp.
explore-lakehouse	Inspect the BigQuery lakehouse — datasets, schemas, sample rows, null rates.
create-dbt-transformations	Build dbt silver/gold models and declare raw sources.
create-cubes	Generate Cube.dev cube definitions from dbt gold models.
query-semantic-model	Run a Cube.js query and render the result inline as a table / chart.
visualize-semantic-model	Render a model-graph.png of the cube relationships.

platform

Do in chat what you'd otherwise do in the RevOS app. Starts with querying your live semantic model in plain English — more of the app's capabilities land here over time. Connects your AI assistant to your org's RevOS data over MCP, registered via .mcp.json.

Skill	What it does
query-semantic-model	Answer a business question by querying the org's semantic model over MCP and rendering the result inline as a table / chart.

Install

skills.sh (npx skills)

skills.sh (the open npx skills tool) installs straight from this GitHub repo. Point it at a bundle directory to install that whole bundle:

# Install the data-engineering bundle into the current project (.claude/skills/)
npx skills add revosai/skills/data-engineering --copy

# List the skills in a bundle without installing
npx skills add revosai/skills/data-engineering --list

# Target a specific agent / install globally / non-interactive
npx skills add revosai/skills/data-engineering -a claude-code -g --copy -y

--copy writes real files (committable, no symlinks). -a targets an agent (claude-code, cursor, opencode, …), -g installs into your user dir, -y skips prompts. Refresh later with npx skills update.

Claude Code (plugin marketplace)

Each bundle is also a plugin in the revos marketplace:

/plugin marketplace add revosai/skills
/plugin install data-engineering@revos

Cowork / claude.ai (plugin marketplace)

Claude chat and Cowork can install straight from this repo's plugin marketplace too, no CLI required:

1. Open Customize in the left sidebar (chat or Cowork).

   

2. Under Personal plugins (or Organization plugins), click + — or Browse plugins if you don't have any yet.

   

3. In the Directory, go to the Plugins tab, click + in the top right, and choose Add from a repository.

   

4. Enter revosai/skills in the URL field (or pick it from the dropdown) and click Sync.

   

5. Switch to the Personal tab — you'll see the skills marketplace listing RevOS Data Engineering and RevOS Platform.

   

6. Click + on the plugin(s) you want to install. Once installed, the + turns into a ⚙️ you can click to open its settings.

   

7. In the plugin's settings, the Skills tab lists the skills it added (e.g. /query-semantic-model).

   

8. For platform, switch to the Connectors tab and click Connect to authorize the RevOS MCP connector — this step is required before the skill can actually query your data.

   

9. This opens RevOS's OAuth screen in your browser. Pick your organization from the dropdown and click Allow. You'll land back in Claude with a "Connected for RevOS" confirmation.

   

10. That's it — ask a business question in plain English and Claude queries your semantic model live:

    Hi! Can you tell me which products are ordered most often?

    

platform — MCP connector

platform is aimed at business users, most of whom aren't in Claude Code at all, so it installs differently — it registers an MCP server, not just skill files, and npx skills add has no defined way to also register a plugin's .mcp.json. Two paths:

Claude Code plugin — registers the MCP server and skill together; OAuth against your RevOS org happens automatically on first tool call, no manual token needed:

/plugin marketplace add revosai/skills
/plugin install platform@revos

Claude.ai / ChatGPT connector (no Claude Code required) — paste the MCP URL directly into the app's connector settings (Claude.ai: Settings → Connectors → Add custom connector; ChatGPT: Connectors → Add MCP server):

https://api.revos.ai/mcp

OAuth happens automatically in-browser the first time a tool is called — no manual token, no CLI, no local files. This path has no skill attached, so the assistant relies on the tool descriptions alone; expect a thinner experience than the plugin path (no chart-rendering convention, no join-ambiguity guidance).

Manual

Each skill is a self-contained folder under its bundle's skills/ directory (e.g. data-engineering/skills/) with a SKILL.md (plus optional references/ and scripts/). Copy a skill folder into the directory your agent watches — e.g. .claude/skills/<name>/ (project) or ~/.claude/skills/<name>/ (global).

Structure

.
├── .claude-plugin/
│   └── marketplace.json       # Claude Code marketplace manifest (one plugin per bundle)
└── <bundle>/                  # e.g. data-engineering/, platform/ (plugin `source`)
    ├── .mcp.json               # optional — MCP server(s) this bundle registers
    └── skills/                 # auto-discovered by Claude Code and skills.sh
        └── <skill-name>/
            ├── SKILL.md        # YAML frontmatter (name, description) + instructions
            ├── references/     # optional, loaded on demand
            └── scripts/        # optional executables

To add a new bundle: create a top-level <bundle>/skills/ directory of skill folders and add a matching plugin entry (with source pointing at the bundle) to marketplace.json.

License

MIT