Generate dbt Validation SQL Notebooks

Generates a SQL Notebook containing validation queries for dbt model or snapshot changes, sourced from either a GitHub PR or a local dbt repository. The notebook opens in the MC Bridge SQL Notebook interface for interactive validation, and includes: two database parameters (prod_db, dev_db) for selecting databases; automatic per-model schema inference from dbt_project.yml and model configs; single-table validation queries; and before/after comparison queries between prod_db and dev_db (or the same database for single-database analysis). Generated SQL uses ANSI-compatible syntax that works across Snowflake, BigQuery, Redshift, and Athena, with minor adjustments sometimes needed for warehouse-specific quirks.

Use it when validating dbt model or snapshot changes with Monte Carlo SQL Notebook queries, from either a GitHub PR or a local dbt repo. Mode is auto-detected from the target argument: a URL or github.com path triggers PR mode, a filesystem path triggers local mode. The skill notes it works well with Sonnet and suggests running /model sonnet before invoking for faster generation.

Required: a target (GitHub PR URL or local dbt repo path). Optional: --mc-base-url <URL> (defaults to https://getmontecarlo.com) and --models <model1,model2,...> to limit which changed models get queries (default: all changed models, up to 10). Prerequisites: the GitHub CLI (gh, authenticated) for PR mode, python3, and the pyyaml package. Two helper scripts do the work: resolve_dbt_schema.py resolves model output schemas from dbt project config, and generate_notebook_url.py encodes the notebook as a base64 import URL and opens it in the browser (<MC_BASE_URL>/notebooks/import#<base64-yaml>).

Data teams running dbt with Monte Carlo who want a generated, ready-to-run validation notebook for a PR's or a local branch's model changes, instead of hand-writing before/after comparison SQL.

Given a GitHub PR URL like https://github.com/monte-carlo-data/dbt/pull/3386, the skill extracts owner, repo, and PR number, then walks through a fixed sequence: fetch PR metadata (gh pr view <PR#> --repo <owner>/<repo> --json number,title,author,mergedAt,headRefOid), fetch the changed-files list, fetch the diff, and filter to only .sql files under models/ or snapshots/ directories at any depth - these are the dbt models actually being validated. If no model SQL files changed, it reports that and stops rather than generating an empty notebook. For each changed model, it fetches the full file content at the head SHA via the GitHub contents API, and separately locates and fetches dbt_project.yml (by finding the common parent directory of the changed files that contains it) for schema resolution.

Notebook cells reference variables via {{param_name}} syntax, defined by dedicated parameter cells. Three parameter types: text (free-form input, e.g. a database name with a default value and placeholder), schema_selector (two linked dropdowns, database then schema, stored as DATABASE.SCHEMA), and dropdown (select from a predefined option list).