Does this write to my warehouse?

Queries are SELECT-only, dry-run first, and capped at 1 GB by default, so the skill reads your warehouse and cannot write to it.

How does it know what a metric means?

It resolves metric names and formulas against your catalog, including 180+ pre-defined calculations, before it writes a query.

Can it query my own tables alongside SourceMedium models?

Queries can join SourceMedium models against your own tables in the same project, with a cardinality check first so a join cannot silently inflate a metric.

What if the CLI does not cover my agent?

Copy the skill folder into that agent's skills directory. It is the same files either way.

Is there MCP support?

MCP protocol support is on the roadmap and is not available today.

Agent Skills

Your coding agent already writes SQL. It does not know which tables are right.

SourceMedium models your commerce data into governed BigQuery tables. These skills teach your coding agent that model, so it queries the right tables and can show you the SQL it ran.

Claude Code
Codex
Cursor
Windsurf

The skills CLI installs into Claude Code, Cursor and Windsurf. Codex and other agents load the same skill folder.

Book a platform walkthrough Read the docs

Already have BigQuery access?

npx skills add source-medium/skills --skill sm-bigquery-analyst

You need Node 18 or newer for the install, the gcloud and bq CLIs, and BigQuery read access provisioned on your SourceMedium project.

They query only the sources connected to your account.

Pointing an agent at raw data is the easy half

An agent with warehouse credentials will happily answer every question you ask. The problem is that you cannot tell which answers are right.

With the skills	On raw data
With the skills The skill discovers your tables and metrics before writing a query	On raw data The agent guesses at table and column names
With the skills Queries run against modeled tables with documented definitions	On raw data SQL runs against raw exports with no documented definitions
With the skills A number you doubt can be traced to the SQL that produced it	On raw data Numbers look plausible and cannot be checked
With the skills Installed once, so a new session does not start from scratch	On raw data Every question starts the discovery process over

“Thanks to SourceMedium, our team no longer wastes time debating metric definitions in meetings. Instead, we can focus on what really matters: making informed, data-driven decisions that drive results.”

Two skills ship today

Install either one on its own. Both read the same modeled tables, so their numbers agree.

SM BigQuery Analyst

Checks your setup, finds the tables and metrics you actually have, then writes SELECT-only SQL you can read before it runs.

Install with --skill sm-bigquery-analyst

Read the skill reference

SM Dashboard Builder

Pins the metric definitions first, checks its own SQL against them, then builds a portable HTML dashboard.

Install with --skill sm-dashboard-builder

Read the skill reference

Questions these skills are built to answer

The analyst skill writes the SQL for the first three. The dashboard skill turns the last one into a page.

“What was my revenue by channel last month?”
“Show me new customer acquisition by source over the past 30 days.”
“What's my customer LTV by cohort?”
“Build an HTML executive dashboard with revenue, orders, AOV, ad spend, and MER for the last 30 days.”

And this is the SQL that comes back with the answer

SELECT
  sm_channel AS channel,
  SUM(order_net_revenue) AS revenue,
  COUNT(DISTINCT sm_order_key) AS orders
FROM `your_project.sm_transformed_v2.obt_orders`
WHERE is_order_sm_valid = TRUE
  AND DATE(order_processed_at_local_datetime) BETWEEN '2024-01-08' AND '2024-01-14'
GROUP BY 1
ORDER BY 2 DESC

SourceMedium's documented example. The line worth reading is is_order_sm_valid: the validity rule lives in the model, so the agent cannot quietly leave it out. How to read a result

Book a platform walkthrough

We will confirm whether your account already has the BigQuery access these skills need.

Before you install

Straight answers, including what is not built yet.

Does this write to my warehouse?: Queries are SELECT-only, dry-run first, and capped at 1 GB by default, so the skill reads your warehouse and cannot write to it.
How does it know what a metric means?: It resolves metric names and formulas against your catalog, including 180+ pre-defined calculations, before it writes a query.
Can it query my own tables alongside SourceMedium models?: Queries can join SourceMedium models against your own tables in the same project, with a cardinality check first so a join cannot silently inflate a metric.
What if the CLI does not cover my agent?: Copy the skill folder into that agent's skills directory. It is the same files either way.
Is there MCP support?: MCP protocol support is on the roadmap and is not available today.

How this fits the rest of the platform

Where this data comes from, and the other ways your team can reach it.

Ready to stop debating the numbers?

In 30 minutes, we will

Map your current systems and reporting gaps.
Show one workflow relevant to your team.
Outline likely fit, implementation scope, and next steps.

Book a platform walkthrough

Share a little context, then choose a time immediately after submitting.

This form requires JavaScript for verification. To request a platform walkthrough, email hello@sourcemedium.com.