SKILL.md
dbt Model Index
When to Use
- Before writing any BigQuery SQL against production data
- When the task has not already explicitly stated which models/tables to query
- When resolving a vague or ambiguous data question into the right BigQuery tables
How to Set Up This Skill
This skill is a curated index of your dbt models. Each entry describes a model (a BigQuery table), what it contains, and what types of questions it is best suited to answer.
To customize this index for your project:
- Organize models into logical domain sections (e.g., Users, Activity, Revenue, Events)
- For each model, include: the table name, a 1–2 sentence description of its grain and content, and "Useful for:" bullets covering common query patterns
- Note key join keys, standard filters, and partition fields where relevant
[Domain: e.g., Users & Identity]
yourmodelname
Brief description of what this model contains. One row per [entity]. Include what makes this model's grain unique and the most important fields.
Useful for:
- [Type of question this model answers, e.g., user counts, cohort sizes]
- [Another use case, e.g., filtering to a specific user segment]
- [Common join pattern, e.g., joining to other tables as the canonical user dimension]
anothermodelname
Description of this model and its grain.
Useful for: [Brief use case description]
[Domain: e.g., Activity & Engagement]
youractivitymodel
Description of the activity signal (e.g., what counts as "active"), the grain, and the time dimension.
Useful for:
- [Use case 1, e.g., daily/weekly active user metrics]
- [Use case 2, e.g., retention analysis]
yourengagementmodel
Description.
Useful for:
- [Use case 1]
- [Use case 2]
[Domain: e.g., Revenue & Subscriptions]
yourrevenuemodel
Description of the revenue grain (e.g., one row per customer per day, or one row per subscription event).
Useful for:
- [Use case 1, e.g., MRR/ARR reporting]
- [Use case 2, e.g., churn analysis]
yoursubscriptionmodel
Description.
Useful for:
- [Use case 1]
- [Use case 2]
[Domain: e.g., Events & Telemetry]
youreventsmodel
Description of the event source, enrichment applied, and key fields available.
Useful for:
- [Use case 1, e.g., raw event-level analysis]
- [Use case 2, e.g., building domain-specific funnels]
Important Notes
- Standard filters: Document any filters that should always be applied in user-facing queries (e.g., excluding test accounts, soft-deleted records, internal users, or flagged/fraudulent users). Example:
where not isinternaluser - Production data: Specify your default project/dataset path. Example:
your-gcp-project.prod.<model_name> - Cost control: For large partitioned tables, always filter on the partition field and constrain the date range to avoid full-table scans
- Model grain: Always note the grain (one row per what?) for each model to avoid accidental fan-outs in joins
- Plan/tier types: If your product has subscription tiers or plan types, document the valid values here so queries filter correctly
- Sensitive datasets: If any models live in a separate dataset, call that out explicitly so queries use the right fully-qualified table reference