Claude Code skills for data product managers. Discovery, validation, and delivery for data products.
A data product is any data asset that someone depends on to make a decision. A risk score that flags patients for readmission. A metrics dashboard that tells a VP which region is underperforming. An API that feeds claims data to a downstream system. If it has consumers, quality expectations, and a reason to exist, it's a data product.
Most data teams ship datasets. A data product operator ships data products — with success metrics defined before data requirements, quality built into pipelines instead of tested in after, and value delivered on a cadence instead of thrown over a wall.
From the Claude Code marketplace:
/plugin marketplace add hollandkevint/data-product-operator
/plugin install data-product-operator@data-product-operator
For local development:
claude --plugin-dir /path/to/data-product-operatorBackground skills activate automatically based on context.
| Skill | What It Does |
|---|---|
data-consumer-discovery |
Discover what internal data consumers actually need. Mom Test for data teams, workaround archaeology, consumer segments, trust barrier assessment. |
data-product-validation |
Score whether a data product idea is worth building. 5-dimension validation scorecard, experiment types, go/investigate/kill thresholds. |
research-synthesis-data |
Convert raw discovery notes into structured insights. Atomic research chain, evidence hierarchy, three synthesis outputs. |
data-team-positioning |
Position data teams as strategic partners, not order-takers. Three stances, evidence as currency, betting table pitch structure. |
data-product-thinking |
Frames problems as data products. Value-first thinking, 5-risk evaluation model, outcome metric trees. |
data-quality-assessment |
Systematic quality evaluation across 5 dimensions: completeness, accuracy, timeliness, consistency, validity. |
stakeholder-alignment |
Translates between technical data teams and business stakeholders. Shapes requests into buildable specs. |
data-model-design |
Star schema, dimensional modeling, OMOP CDM patterns. When to denormalize, how to handle slowly changing dimensions. |
metrics-definition |
Defines metrics with precision: numerator/denominator, grain, time window, edge cases, naming conventions. |
data-team-operating-model |
Squad structure, Shape Up 6-week cycles for data teams, handoff contracts, role assignments by lifecycle stage. |
healthcare-data-domain |
FHIR, OMOP, HL7, clinical terminology (ICD-10, SNOMED, CPT, LOINC, RxNorm), claims data patterns. |
data-storytelling |
Presents data findings to stakeholders. Headline formulas, narrative structures, chart selection, anti-patterns. |
data-pipeline-quality |
Automated quality checks in pipelines. Testing pyramid, dbt patterns, data contracts, circuit breakers, monitoring. |
ethical-risk-assessment |
Bias audit protocols, phased rollout governance, ethics canvas for ML features, HIPAA scope awareness. |
| Command | What It Does | Output |
|---|---|---|
/dpo:run-discovery <idea> |
Guided discovery session: consumer interviews, validation scorecard, go/investigate/kill recommendation | discovery-brief-<name>.md |
/dpo:write-problem-brief <notes> |
Synthesize raw discovery notes into ranked problems with evidence chains and consumer maps | problem-brief-<name>.md |
/dpo:write-data-prd <product> |
Guided PRD creation with data-specific sections: consumer contracts, quality SLAs, schema direction | data-prd-<name>.md |
/dpo:review-data-quality <source> |
5-dimension quality scorecard with scoring rubric and improvement recommendations | quality-review-<name>.md |
/dpo:write-stakeholder-brief |
Translates technical data work into a 1-page business summary with impact metrics | stakeholder-brief-<name>.md |
/dpo:review-data-model <file> |
Evaluates schema design for normalization, naming, relationships, and extensibility | Conversational |
/dpo:reshape-sprint |
Converts sprint artifacts into a shaped pitch: problem, appetite, solution, rabbit holes, no-gos | shaped-pitch-<name>.md |
Discovery to delivery (full chain):
Start with /dpo:run-discovery to interview consumers and score the opportunity. Feed raw notes into /dpo:write-problem-brief to rank problems by evidence. The problem brief feeds /dpo:write-data-prd for requirements. Continue with metrics-definition, /dpo:review-data-model, data-pipeline-quality, and /dpo:write-stakeholder-brief to communicate the plan.
New data product from scratch:
Start with data-product-thinking to scope the problem, then /dpo:write-data-prd to capture requirements. Use metrics-definition to nail down success metrics, /dpo:review-data-model on your schema, data-pipeline-quality to set up tests, then /dpo:write-stakeholder-brief to communicate the plan.
Escape the order-taker trap:
Use data-team-positioning to assess your team's current stance. Run /dpo:run-discovery on your most-used data product to build consumer evidence. Use research-synthesis-data to structure findings into a betting table pitch. Present ranked opportunities backed by evidence, not opinions.
Inherited dataset cleanup:
Run /dpo:review-data-quality to score the current state. Use data-model-design to evaluate the schema, then data-pipeline-quality to add automated checks. Close with data-storytelling to present findings to leadership.
Sprint to Shape Up conversion:
Run /dpo:reshape-sprint on your backlog. The shaped pitch uses data-product-thinking to frame appetite and stakeholder-alignment to translate for the betting table.
Commands work better when they can read your live schemas and sprint artifacts. See CONNECTORS.md for setup instructions. No connections required; every command works with manually provided context.
healthcare-data-domain covers FHIR, OMOP CDM, clinical terminology (ICD-10, SNOMED, CPT, LOINC, RxNorm), and healthcare-specific quality patterns. ethical-risk-assessment adds HIPAA-aware governance, bias testing for clinical models, and phased rollout protocols.
Both activate only when the conversation involves clinical data. Everything else works for any data domain.
Built on the Data Product Operating System (DPOS) — a framework for building data products that actually ship.
- DPOS Blog Series — the thinking behind these practices
- DPOS Assessment — score your data team's maturity across 4 dimensions
Built by Kevin Holland (LinkedIn). 10 years building healthcare data products across 36M patient records.
PRs welcome. If you build data products and have patterns that should be encoded here, open an issue or submit a skill.
Skills follow the conventions in CLAUDE.md: 50-100 lines, imperative voice, concrete examples, zero consultant-speak.
MIT. See LICENSE.
General guidance, not compliance advice. Healthcare skills reflect common patterns but are not a substitute for clinical, legal, or regulatory expertise.