Why

DataDesigner can already discover plugins that are installed in the local Python environment through the data_designer.plugins entry point group. That is the runtime half of the plugin ecosystem.

The missing half is pre-install discovery. Users and the future data-designer plugins ... CLI need a stable tap that can answer what exists, whether it is compatible with the current Python/DataDesigner environment, where the docs live, and what install source should be used. DDPlugins should provide the default NVIDIA tap and serve as the reference implementation for external tap repositories.

System boundary

DDPlugins owns:

• generated catalog metadata
• the default raw JSON tap URL
• plugin packages and per-plugin docs
• scaffolding and reference authoring workflows
• release metadata and validation that keeps the catalog installable

DataDesigner core owns:

• runtime PluginRegistry discovery of installed entry points
• data-designer plugins ... CLI commands
• tap caching and installer execution
• compatibility filtering in the user environment
• post-install entry-point verification

This epic should keep those boundaries clean. DDPlugins should publish a complete tap contract; it should not grow a second runtime plugin loader or installer.

Concrete decisions this stack implements

• Default NVIDIA tap URL: https://raw.githubusercontent.com/NVIDIA-NeMo/DataDesignerPlugins/main/catalog/plugins.json.
• Catalog format: raw JSON, schema version 2.
• Required v2 additions over v1: source and docs.url.
• Source types: pypi, git, and local-authoring-only path.
• Default NVIDIA source type: pypi.
• Generated repo config: [tool.ddp.tap] in the root pyproject.toml.
• Runtime discovery remains installed data_designer.plugins entry points; tap discovery is pre-install metadata only.

Sub-issues

• Define the schema v2 tap catalog contract #16
• Add repo-level tap metadata config for ddp tooling #21
• Generate schema v2 catalog fields from ddp sync catalog #17
• Generate explicit install source metadata for PyPI and Git installs #18
• Publish the generated catalog at a stable raw JSON tap URL #19
• Harden release validation for tap metadata and source refs #20
• Extend ddp new to scaffold all DataDesigner plugin types #22
• Add schema v2 contract fixtures for DataDesigner plugin CLI tests #23
• Clarify first-party plugin tap governance and external tap trust policy #24
• Document the plugin tap contract, workflows, and ecosystem boundaries #25

Dependency order

Acceptance criteria

• catalog/plugins.json is generated as schema v2 raw JSON.
• Every catalog entry includes docs and install source metadata in addition to runtime entry-point and compatibility metadata.
• The default NVIDIA tap URL above is documented and consumable without GitHub API auth.
• Multi-plugin packages list each runtime plugin independently while sharing one package install source.
• Release validation catches stale or incomplete tap metadata before a plugin is published.
• ddp remains the canonical maintainer workflow and becomes usable as a reference tap generator for external repos.
• DDPlugins provides fixtures/docs that DataDesigner can use to implement and test plugins list/search/info/install without DDPlugins-specific special cases.

DataDesigner follow-up markers

These are not DDPlugins work items, but the linked DDPlugins issues should inform future DataDesigner issues:

• DD-CLI-CATALOG-SCHEMA: reject unsupported catalog schema versions and normalize schema v2 entries.
• DD-CLI-DEFAULT-TAP: configure the default NVIDIA tap URL.
• DD-CLI-READONLY-PLUGINS: implement list/search/info/installed behavior.
• DD-CLI-INSTALLER: build install plans from source metadata, prompt before mutation, run uv/pip, and verify entry-point discovery.
• DD-CLI-TAPS: implement external tap add/list/cache/trust UX.
• DD-CORE-PROCESSOR-PLUGIN-VALIDATION: confirm or extend processor validation if scaffolded processor plugins need stronger public validation.