Sigma migration skills

A plugin marketplace of skills for migrating BI tools to Sigma. Each plugin is a pair of skills — a converter (rebuild the content in Sigma) and an assessment (inventory, complexity, and a value/cost shortlist) — validated end-to-end with parity checked against the source warehouse, not just a best-effort port.

Works with any coding agent. Install it as a Claude Code plugin, or use it from Cursor, Cortex Code, and others via AGENTS.md — the skills themselves are agent-neutral (a SKILL.md + scripts/), and credentials load from a shared ~/.sigma-migration/env under any agent.

Install

Claude Code — add the marketplace and install the plugins you need:

/plugin marketplace add twells89/sigma-migration-skills
/plugin install sigma-authoring@sigma-migration-skills      # companion Sigma spec — install alongside any converter
/plugin install tableau-to-sigma@sigma-migration-skills
/plugin install powerbi-to-sigma@sigma-migration-skills
/plugin install qlik-to-sigma@sigma-migration-skills
/plugin install thoughtspot-to-sigma@sigma-migration-skills
/plugin install quicksight-to-sigma@sigma-migration-skills
/plugin install cognos-to-sigma@sigma-migration-skills
/plugin install looker-to-sigma@sigma-migration-skills
/plugin install microstrategy-to-sigma@sigma-migration-skills
/plugin install sisense-to-sigma@sigma-migration-skills
/plugin install gooddata-to-sigma@sigma-migration-skills
/plugin install domo-to-sigma@sigma-migration-skills
/plugin install hex-to-sigma@sigma-migration-skills

Other agents (Cursor, Cortex Code, …) — clone the repo and point your agent at the skill folders; AGENTS.md maps each task to its skill. For example, Cortex Code:

git clone https://github.com/twells89/sigma-migration-skills
cortex skill add sigma-migration-skills/plugins/tableau-to-sigma/skills/tableau-to-sigma

Connect the Sigma MCP server (required). These skills drive your Sigma org through the Sigma MCP server — connect it in your coding agent following Sigma's official guide: Use the Sigma MCP server. It's how the agent reads/queries Sigma (e.g. the parity gate) and builds + validates the migrated data model and workbook. Install it alongside the plugins, before running a migration.

Then just describe what you want migrated — e.g. "migrate this Power BI report to Sigma" — and the skill drives discovery → translation → build → parity.

What's in the marketplace

In Claude Code, installed skills are namespaced — e.g. /powerbi-to-sigma:powerbi-assessment.

The tableau-to-sigma plugin bundles a third skill, tableau-vds-to-cdw — a data-landing bridge for when a Tableau datasource's data lives only inside Tableau (a published extract / VDS feed) and isn't yet in the warehouse Sigma reads. It pulls the data via the VizQL Data Service API and lands it in your cloud warehouse — Snowflake or Databricks, optionally on a schedule — so the converter has a warehouse-native table to build on. tableau-assessment flags the datasources that need it.

The powerbi-to-sigma plugin bundles the equivalent bridge, powerbi-import-to-snowflake — for Import-mode Power BI models whose data lives inside the .pbix / semantic model (Excel, Power Query, flat files) rather than a live warehouse. It reads the model definition via Fabric getDefinition, extracts the stored table rows through the Power BI executeQueries API, and lands them as typed tables in Snowflake — the data track that runs before powerbi-to-sigma converts the model logic. DirectQuery models already point at a warehouse and don't need it.

The shared shape

Every converter follows the same phased flow:

Discover   →  pull the source model + report/sheets/dashboards
Translate  →  measures/calcs/expressions → Sigma formulas (a converter handles the bulk;
              a gap-scout sub-agent validates the hard cases against the live Sigma API)
Data model →  build the Sigma data model (tables + relationships + metrics), reconciled to the warehouse
Workbook   →  rebuild the pages/visuals as a Sigma workbook (layout applied last)
Parity     →  query Sigma vs the source warehouse — GREEN only on a match

Each plugin's skills/<name>/SKILL.md is the entry point; refs/ holds the spec gotchas and scripts/ the pipeline. Skills are self-contained — no external paths to wire up. The canonical phase arc and a per-skill phase-number mapping live in docs/phase-schema.md.

Corpus

corpus/ is a regression corpus: real (demo-data) source artifacts for every tool — .twb, .bim, classic + PBIR report JSON, Qlik app metadata, ThoughtSpot TML, QuickSight describes, Cognos modules/reports, LookML — with golden converter outputs and a runner. Smoke-test converter or builder changes without a live tenant:

corpus/run-corpus.sh --check      # no creds needed; CI-safe

Requirements

• A coding agent that runs skills (Claude Code, Cursor, Cortex Code, …).
• The Sigma MCP server connected in your agent — how the agent reads/queries your Sigma org and builds + validates the migrated data model and workbook. Follow Sigma's setup guide.
• No converter to install — each skill bundles its data-model converter (convert_*_to_sigma) and runs it locally by default (no network, no data egress). A hosted converter MCP is available as an optional, opt-in fallback that sends the source spec off-machine.
• A Sigma API token and a Sigma connection pointing at the same warehouse as the source content (needed for the parity gate).
• Per-tool source access — see each plugin's refs/connection.md (e.g. qlik-cli for Qlik; device-code / Fabric getDefinition for Power BI).

Provenance

The reference migrations and example IDs throughout these skills come from the author's own Sigma + Snowflake test tenant (a retail/workforce star schema). They're included as worked examples — substitute your own connection, data model, and folder IDs.

License

MIT.