Dataproduct Bootstrap

Bootstrap a new dbt data product

Create a new dbt data product project that follows the Entropy Data conventions. This skill handles the greenfield case — empty directory, no dbt project yet. For an existing dbt project that just needs the Entropy Data layer, use the entropy-data-sync skill instead.

What this skill produces

After running, the directory contains:

.
├── dbt_project.yml
├── pyproject.toml
├── .gitignore
├── README.md
├── profiles.yml.example
├── models/
│   ├── input_ports/_models.yml
│   ├── staging/_models.yml
│   ├── intermediate/_models.yml
│   └── output_ports/v1/_models.yml
├── analyses/      # empty
├── macros/        # empty
├── seeds/         # empty
├── snapshots/     # empty
└── tests/         # empty

It then invokes entropy-data-sync to add <id>.odps.yaml, the output-port contract under models/output_ports/v1/<contract>.odcs.yaml, openlineage.yml, and .github/workflows/data-product.yml.

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 1)

Before running Step 1, print this plan to the user verbatim:

Running dataproduct-bootstrap. I'll:

1. Pre-checks: confirm the working directory is empty (greenfield only), then ask whether this is a brand-new data product or one that already has an ODPS draft in Entropy Data.
2. Gather parameters. If you point me at an existing draft, I pull them from the fetched ODPS; otherwise I'll ask you in one batched question (data product id, team, platform, catalog/schema, table).
3. Pick the dbt adapter and profile block for the chosen platform.
4. Scaffold the dbt project (dbt_project.yml, profiles.yml.example, model layout, README, .gitignore), and check whether the user's existing ~/.dbt/profiles.yml would collide with the new profile.
5. Hand off to entropy-data-sync for the publishing layer (ODPS, ODCS, OpenLineage, GitHub Actions).
6. Summarize what was scaffolded and the next manual steps.

Then proceed.

Step 1 — Pre-checks

• Confirm the working directory is empty, or that it contains only files the user is fine with (e.g. an empty git repo, a LICENSE, or a README.md that will be overwritten).
• If dbt_project.yml already exists, stop and tell the user to use the entropy-data-sync skill instead. This skill is for greenfield only.
• Check whether the data product already exists in Entropy Data. Ask the user: "Is there already an ODPS draft for this data product in Entropy Data, or is it entirely new? If existing, paste the data product id or URL." Three outcomes:
  • New — continue to Step 2 with no DATA_PRODUCT preloaded.
  • Existing, id/URL given — extract the trailing id from a URL, then run entropy-data dataproducts get <id> -o yaml. If the lookup succeeds, remember the response as DATA_PRODUCT and use it in Step 2. If it returns a not-found error, tell the user and ask whether to (a) try a different id, (b) proceed as new with that id, or (c) abort.
  • Unsure — if the user does not know, default to asking for the id anyway and run the lookup. Do not silently assume "new".
• CLI prerequisites for the lookup (bootstrap is the one skill where this is global, not via the project venv — see AGENTS.md § Install pattern. The directory has no pyproject.toml yet, so uv run entropy-data is unavailable until Step 4 scaffolds the project and the user runs uv sync. Subsequent skills use uv run entropy-data exclusively.): entropy-data --version must be on PATH (install once with uv tool install entropy-data if missing) and entropy-data connection test must succeed. If either fails, surface the error and ask the user whether to (a) fix the CLI and retry, or (b) skip the lookup and proceed as if new. Don't prompt for the API key yourself; tell the user to run entropy-data connection add <name> --host <host> --api-key <key>.

Step 2 — Gather parameters

Set DBT_PROJECT_NAME = DATA_PRODUCT_ID once DATA_PRODUCT_ID is known.

Step 2a — If DATA_PRODUCT was loaded from Entropy Data

Derive parameters from the fetched ODPS. Treat the draft as authoritative; only ask the user for fields it does not specify.

If the draft declares more than one output port, ask the user which one to use for PLATFORM/CATALOG/SCHEMA/TABLE. Default to the first.

Show the user the derived parameters and ask for confirmation before continuing. Collect any missing fields in one batched question.

Step 2b — If no draft was loaded (new product)

Ask the user for these in a single prompt. Do not generate any files until you have all of them.

Picking TEAM_NAME: prefer a team id that already exists in Entropy Data so the data product slots into the team-scoped views in the UI. If the user does not already know the team id, invoke the entropy-data-teams skill (in this same plugin), let them pick, and use the returned id as TEAM_NAME. A free-text value is still accepted (the ODPS schema does not enforce membership), but the registered id is preferred.

Step 3 — Pick the dbt adapter and profile block

Map PLATFORM to the right dbt adapter package and the profiles.yml.example body:

Step 4 — Scaffold the dbt project

Templates are at ${PLUGIN_ROOT}/skills/dataproduct-bootstrap/templates/. Copy each template into the working directory, substituting placeholders.

Also create empty directories analyses/, macros/, seeds/, snapshots/, tests/. If a directory cannot be empty in git, drop a single .gitkeep file.