| name | entropy-data-publish |
| description | Audit a Declarative Automation Bundle against the Entropy Data reference layout and add anything missing — Open Data Product Specification (ODPS), Open Data Contract Standard (ODCS), and the GitHub Actions publish workflow that deploys the bundle, runs the Lakeflow pipeline, runs the contract test, and publishes ODPS/ODCS back to Entropy Data. Trigger when the user asks to integrate a Databricks bundle with Entropy Data, set up Entropy Data publishing, or check whether a bundle follows the Entropy Data conventions. |
Entropy Data integration for Databricks bundles
Make sure a Databricks Declarative Automation Bundle is well-integrated with Entropy Data.
What "well-integrated" means
A bundle is well-integrated with Entropy Data when it has all of:
| # | Artifact | Path | Purpose |
|---|
| 1 | Open Data Product Specification | <data-product-id>.odps.yaml at repo root | Declares the data product, team, output ports |
| 2 | Output-port data contracts | src/output_ports/v<N>/<contract-id>.odcs.yaml (one per output port — what this data product commits to produce) | Schema + server config the contract test runs against; colocated with the @dp.materialized_view / @dp.table definition that implements it |
| 3 | Input-port data contracts | src/input_ports/<provider-output-port-id>.odcs.yaml (one per active access agreement — what this data product trusts upstream to produce) | Cached snapshot of the upstream provider's ODCS; refreshed via entropy-data datacontracts get, never hand-edited |
| 4 | Bundle layout | src/{input_ports,transformations,output_ports/v1}/ | Convention that mirrors the data product's lifecycle |
| 5 | Publish workflow | .github/workflows/data-product.yml | CI: databricks bundle deploy → bundle run → datacontract test → publish ODPS + output ODCS |
| 6 | Git connections | One per ODPS + one per output-port ODCS, registered via entropy-data dataproducts gitconnection put and entropy-data datacontracts gitconnection put | Lets Entropy Data link the published spec back to the YAML in the repo, and enables pull / push / push-pr from the CLI. Input-port ODCS files are not registered — they belong to the upstream data product, which owns its own git connection |
| 7 | CI lineage emit | .github/workflows/data-product.yml step that calls python3 .github/scripts/emit_lineage.py after bundle run, piping the OpenLineage RunEvent into entropy-data lineage submit. Script lives at .github/scripts/emit_lineage.py | Surfaces pipeline lineage in the platform's Lineage panel. Driven by the CI workflow because lineage edges only change when code/schema changes — daily refreshes don't add edges. Works on serverless pipelines (the lineage-tracking REST API is independent of the pipeline runtime). The script subsets input schemas to the columns the pipeline actually consumes |
How to run this skill
Work in this exact order. Do not skip the audit.
${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 the following plan to the user verbatim so they know what's about to happen:
Running entropy-data-publish. I'll:
- Verify the
entropy-data CLI is installed and connected.
- Confirm this is a Databricks bundle and pick up its name.
- Audit existing Entropy Data artifacts (ODPS, ODCS, bundle layout, publish workflow, git connections).
- Gather any missing parameters from you (one batched question).
- Apply fixes — create missing files, patch incomplete ones, register git connections.
- Summarize what changed and what's deferred.
Then proceed.
Step 0 — Verify the Entropy Data CLI connection
Confirm uv run --quiet entropy-data --version succeeds from the project root. If it fails, run uv sync (the init template seeds entropy-data as a dev dep) and retry. If still missing, stop and tell the user to verify entropy-data is in pyproject.toml's [dependency-groups].dev. Use uv run entropy-data … for every CLI invocation in this skill.
Run entropy-data connection test. If it fails (no connection, expired key, etc.), stop and tell the user to run entropy-data connection add <name> --host <host> --api-key <key> first. Do not prompt for the key yourself.
Step 1 — Confirm this is a Databricks bundle
Check that databricks.yml exists at the working directory root. If not, stop and tell the user this skill only works inside a Declarative Automation Bundle.
Read databricks.yml and remember the bundle.name: value — call it BUNDLE_NAME. By convention it is also the data product id.
Step 2 — Audit
For each row in the table above, check whether the artifact is present. For row 6 (git connections), call:
entropy-data dataproducts gitconnection get <DATA_PRODUCT_ID> -o jsonentropy-data datacontracts gitconnection get <CONTRACT_ID> -o json for each output-port contract under src/output_ports/**/
For rows 2 and 3, glob the file system:
- Output contracts:
src/output_ports/**/*.odcs.yaml
- Input contracts:
src/input_ports/*.odcs.yaml
If a get returns a 404 (or "not found"), mark that connection as missing. If it returns a connection whose repository-url / repository-path / repository-branch does not match the local repo, mark it as drifted and call it out separately — do not silently overwrite. If the underlying data product or contract doesn't exist on the platform yet (the workflow hasn't run for the first time), or if the working directory is not a git repository (git rev-parse --is-inside-work-tree errors or returns false), mark git connections as deferred with a one-line explanation.
For row 1 (ODPS file), also check that the top-level customProperties list contains an entry with property: "dataProductBuilder" and value: "https://github.com/entropy-data/dataproduct-builder-databricks". If the file exists but the property is missing, mark the ODPS as incomplete with a one-line note ("missing dataProductBuilder customProperty"); Step 4 will add it without touching other fields. Forks of this plugin should substitute their own builder URL in the template before publishing.
For row 7 (CI lineage emit), check both files: .github/workflows/data-product.yml must contain a step named Emit lineage to Entropy Data, and .github/scripts/emit_lineage.py must exist. Mark as present when both are wired, missing when either is absent (Step 4c offers to add both), and incomplete when only one of the two is present (likely a partial migration). No per-pipeline configuration is required beyond the existing template placeholders.
Produce a short audit report like:
Entropy Data integration audit for <BUNDLE_NAME>:
[✓] ODPS file
[✗] Output-port contracts (no *.odcs.yaml under src/output_ports/)
[⏸] Input-port contracts (none — populated by dataproduct-implement from access agreements)
[✗] Bundle layout (no src/output_ports)
[✗] GitHub Actions publish workflow
[⏸] Git connections (deferred: data product not yet published — run the workflow first)
Show the report. Then list what you intend to create. Wait for the user to confirm before writing any files.
Step 3 — Gather parameters (only ask for what you cannot infer)
Before generating files, fill in these placeholders. Infer from the project where you can; ask the user for the rest in one batched question.
| Placeholder | Default / inference | Notes |
|---|
DATA_PRODUCT_ID | BUNDLE_NAME | Used as id in ODPS |
DATA_PRODUCT_NAME | Title-cased BUNDLE_NAME | Human-friendly name |
OUTPUT_PORT_NAME | If any src/output_ports/v*/*.odcs.yaml exists, read schema[0].name from it. Otherwise the last segment of BUNDLE_NAME | One output port per ODCS file. The port name should match the table name, not the bundle name |
CONTRACT_ID | If src/output_ports/v*/*.odcs.yaml exists, read its id: field. Otherwise <DATA_PRODUCT_ID>-v1 | Stable id used by entropy-data datacontracts put. When a contract was preloaded from Entropy Data (by dataproduct-init or dataproduct-implement), do NOT mint a fresh <DATA_PRODUCT_ID>-v1 id — use the existing one |
CONTRACT_FILE | If a preloaded ODCS exists, its filename. Otherwise <CONTRACT_ID>.odcs.yaml | File under src/output_ports/v1/ |
CONTRACT_PATH | If a preloaded ODCS exists, its actual path. Otherwise src/output_ports/v1/<CONTRACT_FILE> | Full repo-relative path; used by --repository-path, the CI workflow, and datacontract test |
TABLE | If a preloaded ODCS exists, its schema[0].name. Otherwise the last segment of BUNDLE_NAME | Output table name |
PURPOSE | — | Ask the user (one sentence) |
TEAM_NAME | — | If <DATA_PRODUCT_ID>.odps.yaml already exists with a team.name, use that. Otherwise, prefer a team id registered in Entropy Data — invoke the entropy-data-teams skill (in this same plugin) so the user can pick from the existing teams, and use the returned id. Fall back to a free-text answer only if entropy-data-teams cannot run (CLI unavailable / not authenticated) |
TAG | — | Ask the user (e.g. a usecases/... slug) |
CATALOG | If a single pipeline resource is defined under resources/, take var.catalog from databricks.yml; else ask | Unity Catalog catalog |
SCHEMA | Same — take var.schema; else ask | Unity Catalog schema |
ODPS_FILE | <DATA_PRODUCT_ID>.odps.yaml | Path passed to entropy-data dataproducts put |
API_HOST | entropy-data connection get -o json → host | Resolve in Step 4, only when writing the workflow. Uses the same host the CLI is authenticated against, so CI publish hits the same deployment |
GIT_REPOSITORY_URL | git remote get-url origin | Used by gitconnection put. If no origin, ask the user; if the remote is git@… SSH form, convert to the equivalent HTTPS URL the platform expects |
GIT_REPOSITORY_BRANCH | git rev-parse --abbrev-ref HEAD, falling back to main | Used by gitconnection put; if HEAD is detached, ask the user |
GIT_CONNECTION_TYPE | inferred from GIT_REPOSITORY_URL: github.com → github, gitlab.com → gitlab, bitbucket.org → bitbucket, dev.azure.com / *.visualstudio.com → azuredevops | Ask the user only if the host doesn't match any of these |
GIT_HOST | the URL host, only when self-hosted (i.e. not one of the SaaS hosts above); otherwise omit | Passed as --host to gitconnection put |
GIT_CREDENTIAL_EXTERNAL_ID | — | Optional. Ask the user; if they don't have one yet, leave the connection unauthenticated (it can still be used for read-only metadata in the UI) |
Step 4 — Apply the fixes
For each missing artifact, copy the corresponding template from ${PLUGIN_ROOT}/skills/entropy-data-publish/templates/ into the user's project, substituting placeholders. Do not overwrite existing files; if a file is present but incomplete, surface the diff and ask before changing.
When (and only when) you're about to write .github/workflows/data-product.yml, resolve API_HOST from the active CLI connection:
entropy-data connection get -o json
Use the host field to substitute {{API_HOST}} in the template. Self-hosted deployments are handled via entropy-data connection add --host <host>, not a plugin-level setting.
If the ODPS file exists but was flagged as incomplete — missing dataProductBuilder customProperty in Step 2, append the entry to the top-level customProperties list (do not reorder or touch other entries):
customProperties:
- property: "dataProductBuilder"
value: "https://github.com/entropy-data/dataproduct-builder-databricks"
Surface the diff and ask before saving.
The templates live at:
templates/data-product.odps.yaml → write to <DATA_PRODUCT_ID>.odps.yamltemplates/src/output_ports/v1/contract.odcs.yaml → write to <CONTRACT_PATH> (i.e. src/output_ports/v1/<CONTRACT_FILE>)templates/.github/workflows/data-product.yml → write to .github/workflows/data-product.yml
For the bundle layout, create the directories src/input_ports/, src/transformations/, src/output_ports/v1/ if absent, with .gitkeep files so the directories survive an empty commit. Do not move existing Python files — only add the empty subfolders the user is missing, and note it in the report.
If resources/ does not contain any pipeline resource file, that's likely an unfinished init. Flag it as a separate row ([?] Pipeline resource) and suggest running dataproduct-init or adding a pipeline resource by hand. Do not generate one here.
Step 4b — Configure git connections
Only run this sub-step if the audit (Step 2) flagged at least one git connection as missing or the user confirmed re-creating a drifted one. Skip entirely if every connection is already correct, if the audit marked them as deferred, or if the working directory is not a git repository (check with git rev-parse --is-inside-work-tree — if it errors or returns false, there's no remote to register; tell the user to run git init and add a remote first, then re-run this skill).
For the data product:
entropy-data dataproducts gitconnection put <DATA_PRODUCT_ID> \
--repository-url <GIT_REPOSITORY_URL> \
--repository-path <ODPS_FILE> \
--repository-branch <GIT_REPOSITORY_BRANCH> \
--git-connection-type <GIT_CONNECTION_TYPE> \
[--host <GIT_HOST>] \
[--git-credential-external-id <GIT_CREDENTIAL_EXTERNAL_ID>]
For each output-port ODCS file (src/output_ports/**/*.odcs.yaml):
entropy-data datacontracts gitconnection put <CONTRACT_ID> \
--repository-url <GIT_REPOSITORY_URL> \
--repository-path <CONTRACT_PATH> \
--repository-branch <GIT_REPOSITORY_BRANCH> \
--git-connection-type <GIT_CONNECTION_TYPE> \
[--host <GIT_HOST>] \
[--git-credential-external-id <GIT_CREDENTIAL_EXTERNAL_ID>]
Do not register git connections for input-port ODCS files. They are cached copies of upstream contracts; the upstream data product owns the canonical record.
Notes:
--repository-path is relative to the repo root, not the working directory. The ODPS path is just <DATA_PRODUCT_ID>.odps.yaml; output-port contract paths look like src/output_ports/v<N>/<CONTRACT_FILE>.- Omit
--host for SaaS providers (github.com, gitlab.com, bitbucket.org, dev.azure.com); set it only for self-hosted instances.
- These commands fail if the underlying data product / contract does not exist on the platform yet. If you skipped earlier because of "deferred," surface the manual command in Step 5 so the user can run it after the first workflow run. Do not retry-loop.
- If the audit reported drift (existing connection with different URL/branch/path), confirm with the user before overwriting —
put is upsert.
Step 4c — Add the CI lineage emit step
Only run this sub-step if the audit flagged the CI lineage emit step as missing. Skip if already present.
Ask the user — required confirmation gate:
The CI workflow isn't currently emitting lineage to Entropy Data. Add the step now? It calls Databricks's lineage-tracking API after the pipeline run and POSTs an OpenLineage event for each output port so lineage shows up in the platform's Lineage panel. (yes / no)
If no, mark as skipped in the final report and move on.
If yes, two files need to be added (copy verbatim, substituting placeholders from the audit's resolved values — do not invent either body):
- Workflow step. Insert the
Emit lineage to Entropy Data step after the existing Run Lakeflow pipeline step in .github/workflows/data-product.yml. Source: ${PLUGIN_ROOT}/skills/entropy-data-publish/templates/.github/workflows/data-product.yml. Substitute {{DATA_PRODUCT_ID}}, {{OUTPUT_PORT_NAME}}, {{CATALOG}}, {{SCHEMA}}, {{TABLE}}.
- Script. Copy
${PLUGIN_ROOT}/skills/entropy-data-publish/templates/.github/scripts/emit_lineage.py to .github/scripts/emit_lineage.py in the project. No placeholders — the script reads everything from CLI args supplied by the workflow step.
Step 5 — Final report
Always end with this exact two-part format so the user gets a consistent recap.
Part 1 — outcome table. One row per artifact from the audit. Status uses the shared enum (AGENTS.md § Final-report Status enum). Details is a short, plain-text note (file path, or "—" if nothing to add).
| Artifact | Status | Details |
|---|
| ODPS file | … | … |
| Output-port contracts | … | <N> file(s) at src/output_ports/v<N>/<CONTRACT_FILE> |
| Input-port contracts | … | <N> file(s) at src/input_ports/<provider-output-port-id>.odcs.yaml (or "—" if no access agreements yet) |
| Bundle layout | … | … |
| Publish workflow | … | … |
| Git connections | … | … |
| CI lineage emit step | … | "present" / "added — copied from template" / "skipped — user declined" |
Part 2 — next steps. This skill sits in the middle of the canonical lifecycle (see AGENTS.md § Lifecycle). Include only the items that apply.
The commands below use gh (GitHub CLI) and a .github/workflows/ CI surface — the plugin's default shape. Forks for organizations on GitLab / Azure DevOps / Bitbucket replace gh with the equivalent CLI (glab, az devops, etc.) and swap .github/workflows/data-product.yml for the matching pipeline format.
Pre-first-CI-publish (the data product / contract do not exist on the platform yet):
gh variable set DATABRICKS_HOST --body "<workspace-url>"
gh variable set DATABRICKS_CLIENT_ID --body "<service-principal-app-id>"
gh variable set DATACONTRACT_DATABRICKS_HTTP_PATH --body "/sql/1.0/warehouses/<warehouse-id>"
gh secret set DATABRICKS_CLIENT_SECRET --body "<service-principal-oauth-secret>"
gh secret set ENTROPY_DATA_API_KEY --body "<entropy-data-api-key>"
gh secret set DATACONTRACT_DATABRICKS_TOKEN --body "<sp-or-personal-access-token>"
Post-first-CI-publish (the data product / contract now exist on the platform):
entropy-data dataproducts gitconnection put <DATA_PRODUCT_ID> \
--repository-url <git-url> \
--repository-path <ODPS_FILE> \
--repository-branch <branch> \
--git-connection-type <github|gitlab|bitbucket|azuredevops>
entropy-data datacontracts gitconnection put <CONTRACT_ID> \
--repository-url <git-url> \
--repository-path <CONTRACT_PATH> \
--repository-branch <branch> \
--git-connection-type <github|gitlab|bitbucket|azuredevops>
Other follow-ups:
- "Fill in the data contract schema in
<CONTRACT_PATH> — the template only seeds id and updated_at."
- "Run
dataproduct-implement <data-product-url-or-id> to derive output-port table definitions from the contract."
- "Request access agreements for any upstream input ports —
entropy-data access request <provider-dp-id> <provider-output-port-id> --consumer-dataproduct <DATA_PRODUCT_ID> --purpose <one-sentence>. Once approved, re-run dataproduct-implement to wire them."
If there is nothing in Part 2, write a single line: No further action required.
Conventions and constraints
- No invented schema: when generating the ODCS file, do not invent columns. Seed it with
id + updated_at and tell the user to fill in the rest, or — if @dp.materialized_view / @dp.table definitions already exist for the output port — derive columns from the decorator schema if available.
- Idempotent: running the skill a second time should be a no-op when everything is already present. For git connections that means: if
gitconnection get returns a record matching the local repo URL / branch / path, do not call put.
- Don't overwrite drifted git connections silently. If the platform reports a different URL/branch/path than the local repo, surface the diff and ask. The user may have a fork, a renamed default branch, or a deliberate path remap.
- Don't push secrets: never write API keys, tokens, or workspace URLs into committed files. They must come from GitHub secrets in the workflow.
- Don't create a git repo or commit: leave VCS state to the user.
