bigquery
π―SkillInstallation
npx skills add g1joshi/agent-skills --skill bigqueryRun it in a terminal; it asks which AI coding tools to install it for (Claude Code, Codex, Cursor and more).
Uses skills.sh Β· Node.js 22.20+
After installing
Ask for related work and the agent uses it on its own, or call it yourself.
/bigqueryInstall by hand
Copy the whole skill folder from GitHub to the place below. Copying only SKILL.md can leave out files the skill uses.
Open the skill folder β- This project
.claude/skills/bigquery/ - All projects
~/.claude/skills/bigquery/
These paths are for Claude Code. For other tools, the command above puts it in the right place.
SKILL.md
View on GitHub βExpert Google Cloud BigQuery assistance covering standard SQL, partitioning, clustering, cost optimization, and ML. Use when running petabyte analytics queries, designing data warehouse schemas, or optimizing BI workloads.
Google BigQuery
BigQuery is Google's serverless, highly scalable, and cost-effective multi-cloud data warehouse. It processes terabytes in seconds.
When to Use
- Serverless Enterprise Data Warehousing: Querying petabytes of structured and semi-structured data with zero infrastructure management.
- Real-Time Streaming Analytics: Ingesting and querying millions of events per second with sub-second data freshness.
- In-Database Machine Learning (BigQuery ML): Training and scoring regression, classification, and forecasting models using standard SQL.
- Federated Multi-Cloud Queries (BigQuery Omni): Querying data stored in AWS S3 and Azure Blob Storage directly without egress replication.
Quick Start
-- Standard SQL
SELECT name, COUNT(*) as count
FROM `bigquery-public-data.usa_names.usa_1910_2013`
GROUP BY name
ORDER BY count DESC
LIMIT 10;
Core Concepts
Capacitor Columnar Storage & Dremel Query Engine
Data is stored in proprietary Capacitor columnar format with dynamic tree-based Dremel execution slots:
-- Partitioned & Clustered Table DDL
CREATE OR REPLACE TABLE `my_project.analytics.user_events`
(
event_id STRING,
user_id STRING,
event_name STRING,
event_timestamp TIMESTAMP,
metadata JSON
)
PARTITION BY DATE(event_timestamp)
CLUSTER BY user_id, event_name
OPTIONS(
description="Partitioned clickstream events with clustering for point lookups",
require_partition_filter=true
);
Semi-Structured Native JSON Querying
Queries nested JSON attributes efficiently without schema migrations:
SELECT
user_id,
JSON_VALUE(metadata.device.os) AS os_name,
JSON_EXTRACT_SCALAR(metadata, '$.cart.total') AS cart_total
FROM `my_project.analytics.user_events`
WHERE DATE(event_timestamp) = CURRENT_DATE()
AND JSON_VALUE(metadata.action) = 'checkout';
BigQuery ML Model Training
Trains and evaluates predictive models entirely inside SQL:
CREATE OR REPLACE MODEL `analytics.churn_prediction_model`
OPTIONS(model_type='logistic_reg', input_label_cols=['has_churned']) AS
SELECT
total_spend,
session_count,
has_churned
FROM `analytics.customer_features`;
Common Patterns
Partitioning and Clustering for Cost-Optimized Queries
Problem: Full table scans across billions of event rows lead to astronomical query costs and high latency.
Solution: Partition by date and cluster by high-cardinality lookup keys:
CREATE TABLE `project.analytics.events` (
event_id STRING,
user_id STRING,
event_type STRING,
event_timestamp TIMESTAMP
)
PARTITION BY DATE(event_timestamp)
CLUSTER BY user_id, event_type
OPTIONS (
require_partition_filter = TRUE,
partition_expiration_days = 365
);
Best Practices
Do:
- Always Enforce
require_partition_filter=true: Prevent runaway query billing by requiring users to specify date partitions inWHEREclauses. - Cluster by High-Cardinality Filter Columns: Order tables by user ID, customer ID, or category to minimize scanned bytes.
- Use BigQuery Storage Write API: Stream real-time data using the gRPC-based Storage Write API for reduced cost and guaranteed atomicity.
- Inspect
Bytes Billedin Query Validator: Check dry-run scanned byte estimates before executing heavy exploratory queries.
Don't:
- Use
SELECT *: BigQuery charges per byte scanned; querying all columns drains query budgets. - Use
ORDER BYin subqueries: Global sorting requires single-node processing; sort only in the final outer query withLIMIT. - Export large query results to single CSVs: Use partitioned wildcards (
EXPORT DATA OPTIONS(...)) for multi-part exports.
Troubleshooting
| Error | Cause | Solution |
|---|---|---|
Cannot query without partition filter | require_partition_filter enabled but query lacked date WHERE clause. | Include WHERE DATE(event_timestamp) >= 'YYYY-MM-DD' in query. |
Resources exceeded during query execution | Memory exhaustion from large skew in JOIN or ORDER BY. | Pre-aggregate data, filter early, and avoid ORDER BY without LIMIT. |
Quota exceeded: Exceeded rate limits | Too many concurrent DDL/DML mutation queries per table. | Batch append mutations using BigQuery Storage Write API rather than small single INSERTs. |
References
More from this repository10
A template skill providing a starter framework for creating new agent skills with structured sections for triggers, quick start examples, and capability descriptions.
Skill
Skill
Skill
Skill
Skill
Skill
Skill
Skill
Skill