Pace
All teams / Data

Data

Write SQL, visualize, validate, ship dashboards, interpret without bullshit.

Data work splits roughly three ways: writing SQL, understanding what it returned, and explaining the result. Pace handles all three. Connect your warehouse once and `write-query` learns your schema. Use `validate-analysis` before sharing a conclusion; it has caught Simpson's-paradox-shaped errors more than once.

Install for Data

The full data starter set is 3 plugins. Pick the install path that matches you.

Cowork (no terminal)

Click each link below; Cowork opens and asks you to confirm. Restart Cowork after the last one so connectors register.

  1. 1 Install data
  2. 2 Install engineering
  3. 3 Install finance
After the last install, restart Cowork. The MCP servers (Slack, HubSpot, etc.) only register cleanly on a fresh app start. Quit the app fully, then reopen it.

Terminal (one command)

If you have claude on your PATH already, this installs the whole set in one go.

npx pace-tools install data engineering finance

pace-tools wraps claude plugin install; runs claude plugin marketplace add GoldenBerry-SO/Pace automatically if you haven't registered the marketplace yet.

Plugins to install

Pick from this set for the role. Primary plugins are essential; companions multiply value.

  • data primary

    SQL drafting, visualization, statistical validation, dashboards, anomaly interpretation.

    claude plugin install data@pace
  • engineering companion

    When data questions overflow into pipeline or schema territory.

    claude plugin install engineering@pace
  • finance companion

    Variance analysis and financial-system queries lean on data muscle.

    claude plugin install finance@pace

Connectors to set up

Claude prompts to authorize each one the first time a relevant skill fires. You only do this once per project.

  • Snowflake
  • Databricks
  • BigQuery
  • Hex
  • Amplitude
  • Jira

Workflows

Common ways teams use these plugins day to day. Each one is a starting point; adapt the prompt to your context.

  • Write SQL from intent

    Describe what you want; get SQL back, sized to your schema. Connecting a warehouse means it can also run it and return results.

    Write a SQL query for monthly active users by plan tier, last 6 months.

  • Visualize results

    Suggests the right chart type for a result set and generates the visualization (Hex notebook, plot code, or chart spec).

    Visualize this query result. Pick the chart type that tells the story.

  • Validate an analysis

    Pre-flight check before sharing a conclusion. Looks for confounds, sampling issues, base-rate mistakes, Simpson's paradox shapes.

    Pressure-test my conclusion that conversion improved 12% in Q3. What am I missing?

  • Statistical test

    Runs the right test (chi-square, t-test, mann-whitney) and reports significance, effect size, and what it actually means in business terms.

    Run the right statistical test on these A/B results from last sprint.

  • Build a dashboard

    Generates a dashboard spec (Hex, Looker, or Mode-compatible) from a set of queries.

    Build a weekly product-health dashboard. Pick the metrics that matter.

  • Interpret an anomaly

    Given a chart or metric drop, generates plausible explanations ranked by likelihood and what data would distinguish them.

    DAU dropped 8% last Tuesday. Interpret it. Real or noise?

Just say it

You don't have to memorize slash commands. After installing, type natural sentences into Claude and the right skill will fire. Each row is a real example.

  • You say

    Write a SQL query for revenue by region last quarter

    Claude triggers write-query
  • You say

    Analyze churn over the past 90 days

    Claude triggers analyze
  • You say

    Are there anomalies in our signup funnel?

    Claude triggers statistical-analysis
  • You say

    Build a dashboard for product engagement

    Claude triggers build-dashboard
  • You say

    Visualize this query result as a chart

    Claude triggers create-viz
  • You say

    Sanity-check this dataset before I share it

    Claude triggers validate-data

Tips & tricks

Field-tested patterns from teams that have already shipped a quarter or two on these plugins.

  • Always validate before sharing

    `validate-analysis` catches the errors that get spotted by the audience in the meeting. Cheap insurance.

  • Connect the warehouse first

    Without a connector, the data plugin writes SQL but can't run it. With one, the workflow is interactive.

  • Document queries you keep

    `document-query` before saving SQL to your team library. Future-you appreciates context.

  • Use the three-step rhythm

    For exploratory: `write-query` → `visualize` → `interpret`. Quick, surprisingly thorough.

  • Pair with engineering for perf

    When the query is slow, hand off via `code-review` of the EXPLAIN plan.

  • Statistical tests are not optional for A/B

    `statistical-test` instead of eyeballing two means. It tells you whether the effect is real.

See also