Dataproduct Exampledata

Upload example data for a data product

Sample rows let prospective consumers evaluate a data product without requesting access. This skill pulls a small sample from a non-production source, scrubs sensitive columns, and uploads the result.

How to run this skill

${PLUGIN_ROOT} below refers to the root of this plugin — the directory that contains skills/. On Claude Code it is set automatically as ${CLAUDE_PLUGIN_ROOT} — use that. On any other agent (Codex, Copilot CLI, etc.) it is unset; resolve it as ../.. relative to this SKILL.md file's directory (i.e. the grandparent of skills/<this-skill>/).

Plan announcement (before Step 0)

Before running Step 0, print this plan to the user verbatim:

Running dataproduct-exampledata. I'll:

1. Pre-checks: dbt project, ODCS files, entropy-data CLI, non-prod dbt target.
2. Identify the output port and its contract.
3. Build a scrub plan — drop PII/sensitive columns, hash IDs, drop free text. Wait for your confirmation.
4. Extract ~20 sample rows via dbt show against the non-prod target.
5. Build the example-data YAML and show the first rows. Wait for your confirmation.
6. Upload via entropy-data example-data put.
7. Summarize what was uploaded, what was scrubbed, and cleanup options.

Then proceed.

Step 0 — Pre-checks

• Confirm dbt_project.yml exists at the working directory root.
• Confirm there is at least one output-port ODCS file under models/output_ports/.
• Confirm uv run --quiet entropy-data --version succeeds from the project root. If it fails, run uv sync and retry; if still missing, stop and tell the user to verify entropy-data is listed in pyproject.toml's [dependency-groups].dev. Use uv run entropy-data … for every CLI invocation in this skill.
• Confirm a non-production dbt profile/target exists (test, dev, or similar). Inspect profiles.yml if accessible; otherwise ask. Never use a prod target in this skill.

Step 1 — Identify the output port

If multiple output ports exist, ask which one. For each candidate, you need:

• OUTPUT_PORT_ID (from <id>.odps.yaml)
• the matching models/output_ports/v<N>/<file>.odcs.yaml
• the table name and server config the contract points at

Step 2 — Build the scrub plan

Read the contract's field list. For each field, decide what to do with it:

Show the user the scrub plan as a table — column → action — and wait for confirmation before extracting any data.

Step 3 — Extract the sample

Build the SQL:

select <kept-and-hashed-columns>
from <contract-server-table>
limit <N>;

Default N = 20. Use the dbt non-prod target chosen in Step 0. Preferred extraction methods, in order:

1. dbt show --inline "<sql>" --target <non-prod-target> — uses the dbt connection, no extra credentials needed.
2. As a fallback, ask the user to run the query themselves and paste the result.

Convert the result rows into a list of objects keyed by the contract column names (the names that will be visible to consumers, not the warehouse aliases). Hold the rows in memory as ROWS for the next step — do not write a CSV. The entropy-data example-data put command takes a YAML/JSON body, not a CSV.

Step 4 — Build the example-data document and show the sample

Construct the document the CLI expects:

id: <DATA_PRODUCT_ID>-<OUTPUT_PORT_ID>
dataProductId: <DATA_PRODUCT_ID>
outputPortId: <OUTPUT_PORT_ID>
dataContractId: <CONTRACT_ID>
schemaName: <model name from the ODCS `models:` block>
data:
  - { <col>: <val>, ... }   # one entry per row from ROWS
  - ...

Field semantics confirmed against entropy-data example-data list -o json: the ID convention is <dataProductId>-<outputPortId>; schemaName is the contract's top-level models: key (the table name as the contract names it).

Write the document to examples/<DATA_PRODUCT_ID>-<OUTPUT_PORT_ID>.yaml (create examples/ if missing; add examples/ to .gitignore if absent).

Print the first 5 rows of data: in a Markdown table. Re-state the dropped columns. Wait for explicit user confirmation before uploading.