v0.6.0

v0.6.0 - 2026-06-10

Added

• After setup, dlthub-start automatically runs your first pipeline on dltHub: it logs in, binds the project to a playground workspace (creating it only if it doesn't already exist), runs load_sample_shop with --follow (waiting for the run to finish), and opens the workspace overview with dlthub show. The post-run next-steps panel then shows a ready-to-paste prompt addressed to your chosen coding agent (claude/cursor/codex) for building a pipeline against your own source — and copies it to your clipboard so it's one paste away. Skipped under --yes, since the login is interactive.

Changed

• uv sync and the first pipeline run now happen automatically instead of being prompted. --skip-uv-sync still opts out of dependency sync (and, with it, the first run).
• dlthub-start now initializes a workspace in place — the current directory by default (the workspace-name prompt is gone). An explicit positional argument still targets a named subdirectory. Either way the target must be empty; a non-empty target stops with a dedicated "Directory not empty" message and exit code 2 instead of a generic error. Next-steps guidance adapts: the cd step is omitted when initializing in place, and a note reminds AI agents to run from the workspace root when a subdirectory is used.
• Bundled AI toolkits are now one-shot-pipeline and dlthub-platform (previously data-exploration, dlthub-platform, and rest-api-pipeline).
• The minimal workspace now also depends on pyarrow and ibis-framework[duckdb] for local data access.
• Bumped WORKBENCH_REF to 21b9bba4 and refreshed the bundled AI workbench scaffolds.
• make check-ai is clearer: it confirms when the bundled scaffolds are up to date, lists exactly which files differ (including untracked additions) when they're not, and hides the noisy generate-ai output unless it fails.

Removed

• Removed the starter_workspace scaffold along with the --scaffold flag and the interactive scaffold picker. minimal_workspace is now the only bundled scaffold.
• The minimal workspace no longer ships report_notebook.py. On dltHub Platform the notebook had to re-run the whole pipeline to repopulate the worker-local DuckDB before charting, which made it slow and undercut the first-run experience. It's removed temporarily until shared workspace storage lands and the notebook speed issue is fixed; agents can still build notebooks on demand via the data-exploration toolkit.

Fixed

• python -m create_dlthub_workspace now propagates the CLI exit code (it previously always exited 0 because __main__ ignored main()'s return value).