| name | raincloud-remove-dataset |
| description | Remove a dataset from sources.json and clean up its outputs. Use when the user wants to delete a slug from the manifest. Destructive — always confirm before acting. |
| argument-hint | <slug> |
| disable-model-invocation | true |
Walk through removing dataset $ARGUMENTS. Reference: SKILLS.md "Removing a dataset".
This is destructive. Confirm with the user before doing any of these steps. No backwards-compat shims are kept — git history is the fallback (.archive/ is gitignored and only present on the maintainer's tree).
Steps:
-
Remove the DatasetSpec entry from sources.json using the Python load-edit-dump pattern — never sed. Look up by slug and remove from m["datasets"].
-
Delete the output parquet (and sibling .vortex if present):
rm -rf outputs/v1/<slug>/
-
Optionally delete the raw cache:
rm -rf outputs/raw_downloads/<slug>/
Only do this if you're confident the dataset won't be re-added. The raw cache is unversioned and could feed a future schema_version bump.
Watch for sibling slugs sharing the same upstream URL (GloVe sizes, OSM Germany kinds) — those are deduped via hardlink. Removing outputs/raw_downloads/<slug>/ is fine because each slug gets its own subdir, but verify there isn't a multi-output handler producing the parquet from a shared raw dir before deleting either.
-
Regenerate docs:
python -m scripts.pipeline.docs
(Or invoke /raincloud-docs.)
-
If a handler became unused (only this slug referenced it), remove the handler file from scripts/pipeline/handlers/ and unregister it from __init__.py. No stub or deprecation shim — fully delete.
Removed means removed.
