snowflake-clone-notebook

> **Paths.** `<plugin-root>` is this plugin's directory, two levels above > this `SKILL.md`. Write its absolute path wherever `<plugin-root>` appears. # Standard-catalog table-creation script **This is the Standard-catalog path, and only for a Standard catalog the user explicitly asked for.** The default target is an EXTERNAL/SNOWFLAKE catalog, which needs no tables at all — see `snowflake-medallion-clone` Phase A. Do not reach for this skill just because a migration is in progress. Why a script on compute: it runs **inside AIDP compute**, so every statement's success, failure and elapsed time appears in the cluster's own output, and each object is read back. The control-plane catalog API is used for the catalog container only. Generate locally. `notebook --upload` is a dry run and is refused with `--execute`; the structure is created by `run --job snowmig_01_structure` (S10). **The script creates empty structure and moves no data.** Every table it creates has its columns and zero rows. A user who hears "clone" may expect rows — say this before they run it. ## 1. Generate (offline, safe) ```bash "<plugin-root>/bin/snowmig" notebook \ [--catalog <catalog>] ``` Writes `snowmig_shallow_clone_<catalog>.ipynb` locally plus `NOTEBOOK.md`. One notebook per catalog — bronze mirrors the source, so a multi-database estate has several. ## 2. `--upload` — a dry run, refused with `--execute` ```bash "<plugin-root>/bin/snowmig" notebook \ --upload --datalake-ocid <ocid> --workspace <ws> --cluster-id <cl> --catalog <cat> ``` `--upload` is a **dry run** without `--execute`: it prints where the notebook would land and sends nothing. With `--execute` the upload is **refused**, and the command says so. The structure is created by `snowmig provision --execute` (which places the stage notebooks in the workspace) followed by `snowmig run --job snowmig_01_structure` (S10). Ask the user for the four coordinates in this turn; nothing is stored. Nothing lands there from this command. `NOTEBOOK.md` records `/Workspace/Shared/snowmig_shallow_clone_<catalog>.ipynb` as the intended path, and if the user places it there themselves from the workspace UI, the **shared** directory is the right home — re-runnable, readable and debuggable independently of this plugin and of the conversation that generated it. **Notebooks live in the workspace filesystem, not in a data catalog** — catalogs hold tables and views. Say that if the user expects to find it under a catalog. ## 3. Execution is the user's call **Do not run it for them.** Ask, then let them run it in the AIDP workspace after they agree. The notebook is built to be watched: each object prints `[3/7] D.S.ORDERS ... ok (0.4s)`. If a run is slow, that output is where the progress is — report the last line rather than guessing. The final cell prints `verified N/M` plus any missing object. ## What the notebook does and does not do - Creates schemas → tables → views, in that order, views after the tables they read - `IF NOT EXISTS` throughout, so re-running is safe - **No `INSERT`, `COPY INTO`, `MERGE`, `UPDATE`, `DELETE`, `TRUNCATE` or `DROP`.** Its metadata records `moves_data: false` - Tables arrive **empty**. Say this out loud — a user seeing "clone" may expect rows - Blocked objects appear in markdown only, never in a code cell, so it cannot attempt them