Przejdź do treści
Biblioteka skilli
Dane i analityka · Budowanie i kod

dbt-bigquery

/dbt-bigquery

Tworzy, zmienia i testuje modele dbt na BigQuery: warstwy, strategie przyrostowe, snapshoty SCD2, partycjonowanie i kontrakty.

Repozytorium
wraith-skills ↗
Licencja
MIT
Ostatnia zmiana repozytorium
13.11.2026

Ocena i bezpieczeństwo

90/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść25 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
4
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Tworzy, zmienia i testuje modele dbt na BigQuery: warstwy, strategie przyrostowe, snapshoty SCD2, partycjonowanie i kontrakty.

Oryginalny opis (EN)

Use when creating, changing, testing or running dbt models on BigQuery: layering, incremental strategies, ds snapshots, SCD2, partitioning and clustering, sources and freshness, contracts and safe runs against dev targets.

Jak działa

  1. When to use
  2. Rules
  3. Workflow
  4. Commands
  5. dbt Labs skills
  6. Review checklist
  7. When it fails
  8. References
Sekcje z nagłówków SKILL.md, w oryginale · 98 linii, 838 słów
Podgląd SKILL.md
---
name: dbt-bigquery
description: "Use when creating, changing, testing or running dbt models on BigQuery: layering, incremental strategies, ds snapshots, SCD2, partitioning and clustering, sources and freshness, contracts and safe runs against dev targets."
license: MIT
metadata:
  source: new
---

# dbt on BigQuery

## When to use

Creating, changing, testing or running dbt models on BigQuery.
Models are built from an ODCS contract (wraith:odcs-contracts).
Project values come from the `## wraith` block of AGENTS.md.

## Rules

1. dbt is a build system for SQL. The commitment is to the contract, not to dbt.
2. Layers follow the work:
   - raw: source-aligned, generated from the contract, incremental merge with dedup;
   - staging: typed, renamed to contract names, a `ds` snapshot;
   - intermediate: only when logic is reused or complex;
   - marts: consumer-facing, contract enforced.
3. Datestamp first: snapshot models are partitioned on a DATE column `ds`. Each run writes exactly one `ds` with `incremental_strategy='insert_overwrite'`. The value comes from `{{ var('ds') }}` set by the orchestrator, so a backfill is a rerun with an earlier `ds`.
4. SCD2 only for regulated audit trails. Build it by full rebuild from the `ds` snapshots with LEAD over `ds`; never a hand-rolled merge, and no dbt snapshots for new work.
5. Uniqueness on snapshots is tested on `(key, ds)`.
6. Raw dedup uses `incremental_strategy='merge'`, with `unique_key` set to the contract primary key and `row_number()` on the loaded-at column. Read new rows with a lookback window, never a strict greater-than high-water mark.
7. Always set `partition_by` (DATE or TIMESTAMP, never BYTES) and `cluster_by` (at most 4 columns used in filters and joins).
8. Incremental models set `on_schema_change='fail'`.
9. Full rebuild (`table`) until scan cost is measurably too high; state the cost threshold in the model's description. `dbt run --full-refresh` is the migration story for anything rebuildable from source. Raw tables that cannot be rebuilt set `full_refresh: false`.
10. Marts set `contract: {enforced: true}` with a `data_type` per column.
11. Tests come from the contract quality rules. Add dbt unit tests for non-trivial logic. Every source with an SLA has freshness configured.
12. Orphans surface as NULLs (LEFT JOIN to optional relations). A silently dropped row is invisible corruption.
13. Ratios use `SAFE_DIVIDE`.
14. A metric is defined in one model; dashboards read it and never redefine it.
15. Build at the most granular meaningful grain and aggregate late.
16. Agents run dbt only against `dbt_target_dev`, with `maximum_bytes_billed` set to `dev_max_bytes_billed` in that profile. `dbt_target_prod` runs happen only through wraith:deploy, with the user's go-ahead.
17. Run dbt from `dbt_project_dir`. `target/`, `logs/` and `dbt_packages/` are gitignored. CI runs `dbt deps` before `dbt build`.
18. When scanning dbt output, match `ERROR creating|ERROR in|Database Error|Compilation Error`, never bare `ERROR`.
19. After changing an extract, run `dbt test --select source:<name>` on a real sample before any full rebuild.
20. No hardcoded date literals in models; use vars or `ds`.

## Workflow

1. Read the `## wraith` block of AGENTS.md. If `dbt_project_dir`,
   `dbt_target_dev`, `dbt_target_prod`, `dev_max_bytes_billed`,
   `gcp_project_dev`, `bigquery_location` or `contracts_dir` is missing, ask
   once and suggest wraith:setup.
2. Read the contract in `<contracts_dir>`; mapping to models and tests is in
   wraith:odcs-contracts.
3. Write the test first, following `references/data-tests.md`:
   contract-derived tests and a unit test for non-trivial logic.
4. Write the model in its layer using the patterns in `references/patterns.md`.
5. Run against `<dbt_target_dev>` only, from `<dbt_project_dir>`.
6. Scan output with the rule 18 pattern; fix and rerun.
7. Production runs go through wraith:deploy with the user's go-ahead.

## Commands

All run from `<dbt_project_dir>`. `dbt deps` downloads packages from the
network; the agent runs it visibly, never from a hook.

- `dbt deps`
- `dbt build --target <dbt_target_dev> --select <model>+ --vars '{ds: <YYYY-MM-DD>}'`
- `dbt test --target <dbt_target_dev> --select source:<name>`
- `dbt source freshness --target <dbt_target_dev>`
- `dbt run --target <dbt_target_dev> --select <model> --full-refresh` (rule 9 only)

## dbt Labs skills

When dbt Labs' dbt plugin is installed, use its skills for generic dbt mechanics:

- `running-dbt-commands`: any dbt command run from a wraith skill.
- `using-dbt-for-analytics-engineering`: everyday dbt model work.
- `adding-dbt-unit-test`: test-first work on a dbt model, beside `references/data-tests.md`.
- `maintaining-dbt-documentation`: backfilling model and column descriptions, with review.

Where they and this skill disagree, this skill wins for contract-first models, ds snapshots and SCD2, BigQuery cost guards and production safety (rules 3, 4, 9 and 16).

## Review checklist

- grain stated in the model description
- no `SELECT *` in models
- partition filter on large sources
- `SAFE_DIVIDE` for ratios
- optional relations use LEFT JOIN, so orphans surface as NULL
- incremental models set `on_schema_change: fail`
- no hardcoded date literals (use vars or `ds`)
- a test for every contract quality rule
- marts have the contract enforced

## When it fails

- Compile errors happen before any SQL runs; runtime errors come from the warehouse. Read `target/run_results.json` and `logs/dbt.log` for the failing node.
- Match `ERROR creating|ERROR in|Database Error|Compilation Error`, never bare `ERROR`: the summary line prints `ERROR=0` on success.
- Connection problems: run `dbt debug`.
- Rerun just the failing model with `--select <model>` to keep the loop fast.

## References

- [Patterns](references/patterns.md): raw merge, `ds` snapshot, SCD2 rebuild, dev profile.
- [CI](references/ci.md): `dbt deps`, slim build, Workload Identity Federation.
- [Data tests](references/data-tests.md): generic tests from contract rules, unit tests, golden files, backtests.
Pliki w paczce (4)
  • SKILL.md 6 KB
  • references/ci.md 1 KB
  • references/data-tests.md 2 KB
  • references/patterns.md 2 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · wraith-proxy/wraith-skills90MIT
gemini-cli-extensions/data-agent-kit-starter-pack100Apache-2.0
420company/artemis95MIT

Migawka bazy z 13.11.2026 · wersja paczki 9ed6347923d1 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 1 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 90data-platform
    Wspiera lokalne w repozytorium przepływy BigQuery, Databricks i dbt, z ograniczonymi zasadami bezpieczeństwa BigQuery i dbt oraz uwierzytelnianiem w chmurze.
  • 100dbt-migration-bigquery
    Konwertuje DDL Google BigQuery na modele dbt zgodne ze Snowflake: widoki, tabele i procedury składowane, z plikami schema.yml, testami i dokumentacją.
  • 100dbt-migration-bigquery
    Konwertuje DDL Google BigQuery na modele dbt zgodne ze Snowflake: widoki, tabele i procedury składowane, z plikami schema.yml, testami i dokumentacją.
  • 95dbt-transformation-patterns
    Opisuje pracę z dbt w inżynierii analitycznej: organizację modeli, testy, dokumentację i strategie przyrostowe.
  • 100dbt-transformation-patterns
    Obejmuje dbt w inżynierii analitycznej: organizacja modeli, testowanie, dokumentacja i strategie przyrostowe przy budowie transformacji danych.
  • 100building-dbt-models
    Buduje dobrze ustrukturyzowane modele dbt: warstwy staging, intermediate i marts, ref() i source(), materializacje oraz modele przyrostowe z właściwą strategią.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.