Consommé — Clarification

Now part of Trousse. The consomme skill was absorbed into trousse in v0.5.0. Install trousse and the analysis skill comes with it. The standalone spm1001/consomme repo is archived.

BigQuery data analysis for AI coding agents. Messy data goes in, crystal-clear insights come out — like the classical technique it’s named for, where a raft of egg whites draws impurities from a murky stock and leaves behind a perfectly transparent broth. The technique does the work, not the person.

Consommé is a skill that pairs Google’s BigQuery Data Analytics MCP extension with a structured methodology adapted from Anthropic’s data plugin approach. It gives Claude Code a repeatable workflow from “what tables do we have?” to an interactive HTML dashboard with KPI cards and filters.

How to get it

Install trousse — consomme comes with it:

claude plugin install batterie-de-savoir/trousse

The skill is invoked as /consomme (or trousse:consomme). Seven commands provide shortcuts: /consomme-profile, /consomme-explore, /consomme-dashboard, /consomme-validate, /consomme-sheets, /consomme-ingest.

When to use / When NOT to use

Use Consommé when:

You need to explore, query, or analyse structured data in BigQuery
You want interactive visualisations — HTML dashboards with Chart.js, filters, KPI cards
You’re building funnels, cohort analyses, or forecasts from warehouse data
You need the agent to profile a dataset before jumping to conclusions

Do NOT use Consommé when:

You need content from Google Workspace (Docs, Sheets, Gmail, Drive) — use Mise
You need to prepare raw data for BigQuery (SPSS, codebooks, schema design) — use /mandoline (also in trousse)
You’re working with local CSV/Excel files that aren’t in BigQuery
You need real-time streaming data rather than analytical queries

Key concepts

5-stage methodology

Discovery — Catalogue search. What datasets and tables exist? What’s the schema?
Exploration — 3-phase data profiling: structural (row counts, partitioning), column-level (types, nulls, cardinality), and relationships (keys, joins, referential patterns).
SQL craft — BigQuery-specific SQL reference: window functions, CTEs, funnels, cohort analysis, approximate aggregation.
Analysis — Execute queries via MCP tools (execute_sql, forecast, analyze_contribution). Interpret results in context.
Validation — Pre-delivery QA framework. Sense-check results before presenting.

Companion skill: Mandoline

Mandoline (/mandoline, also in trousse) is the upstream counterpart — it transforms raw data (SPSS files, messy spreadsheets) into self-documenting BigQuery tables. Consommé analyses what mandoline produces. Together they cover the full pipeline from raw survey data to interactive dashboards.

How it relates to other tools

Tool	Relationship
Mise	Mise handles Google Workspace content. Consommé handles structured data in BigQuery. The boundary is explicit — consommé’s skill routes Workspace requests to Mise.
Mandoline	Data prep (in trousse). Transforms raw data INTO BigQuery. Consommé analyses data already IN BigQuery.
Bon	An analysis task tracked as a bon outcome; consommé does the analytical work within that outcome.
Garde-manger	Past analyses are searchable in garde-manger.

Source

The skill lives in trousse at skills/consomme/. The archived standalone repo is at spm1001/consomme.