DataDesigner Repo Skill

Use this skill for the DataDesigner repository and its published Python packages. It is a router, not a full manual.

Start here

• Read references/package-overview.md for the package layout, verified entry points, and route map.
• Read references/troubleshooting.md when the package or CLI fails to import, model aliases are missing, persona assets are absent, or a generation path needs external credentials.
• Read references/repo-provenance.md when you want to check whether this skill still matches the current checkout.
• Run scripts/check_datadesigner_environment.py if you want a quick read-only smoke check of the active installation.

Install and smoke check

For a published install, use:

pip install data-designer

For a workspace checkout, use the repo's uv workspace sync instead of hand-installing individual packages:

uv sync --all-packages --no-default-groups

Minimal import check:

python -I -c "import data_designer.config as dd; from data_designer.interface import DataDesigner; print(dd.DataDesignerConfigBuilder, DataDesigner)"

If data-designer agent context reports no usable model aliases, that is a configuration state, not an install failure. Config-only workflows can still validate.

Route map

config-authoring

Use for dataset schema design: builder API, columns, samplers, validators, processors, seed sources, custom columns, person data, and config troubleshooting.

generation-runtime

Use for DataDesigner.validate, preview, create, check_models, set_run_config, result handling, export, push to Hugging Face Hub, resume, and workflow chaining.

cli-and-agent-tools

Use for the data-designer CLI, config/model/provider commands, download/persona management, plugin catalog commands, and data-designer agent context/types/state.

plugins-and-extensions

Use for custom column generators, plugin entry points, plugin install/uninstall planning, MCP tool configs, and installed-plugin inspection.

recipes-and-integrations

Use for notebook tutorials, code/SQL/image/VLM recipes, trace ingestion, workflow chaining, human review flows, and Hugging Face export guidance.

What not to do

• Do not depend on files from the original checkout at runtime. Use only bundled references and scripts under this skill tree.
• Do not tell future agents to open original repo docs or examples when a bundled reference exists.
• Do not treat optional remote model, persona-download, GPU, or Docker flows as required for the base package unless the task explicitly asks for them.

Read next

• references/package-overview.md for the verified package map and workflow groups.
• references/troubleshooting.md for cross-cutting failures.
• sub-skills/<id>/SKILL.md for the specific workflow you need.