query

name: query description: Runs a read-only SQL query against the Customer Experience tables (mozdata.customer_experience) and returns results as a formatted table. Use when the question requires counts, aggregations, rankings, or distributions — not when it requires reading document text.

Query Skill

Runs a read-only SQL query against BigQuery and prints results as a formatted text table. No embedding required.

This skill is locked to mozdata.customer_experience — it can read any table or view in that dataset, and anything outside it is rejected. Only a single read-only SELECT (optionally WITH … SELECT) is accepted. The dataset boundary is enforced by asking BigQuery (via a dry run) which tables the query actually reads, so it can't be evaded by SQL the script doesn't parse.

Authentication

This skill connects to BigQuery using a service account — read-only access is enforced on the impersonated token. Just log in:

gcloud auth application-default login

Your authenticated BigQuery credentials are only used to create a new and temporary read-only access token on behalf of the service account defined in SERVICE_ACCOUNT, which requires the roles/iam.serviceAccountTokenCreator role on that service account. Your login no longer needs --scopes.

Usage

python ${CLAUDE_PLUGIN_ROOT}/skills/query/scripts/query.py \
  --sql "<read-only SELECT over mozdata.customer_experience.<table>>"

Reference tables by their fully-qualified name (mozdata.customer_experience.<table>).

When to use

Use this skill instead of embed + vector-search when the question requires:

• Counts ("how many tickets about X?")
• Rankings ("top N topics by volume")
• Distributions ("ticket count by product")
• Aggregations ("average sentiment by category")
• Any answer that is a number, ranked list, or summary statistic

Use embed + vector-search when the question requires reading and synthesizing document text.

Common SQL patterns

Dataset: mozdata.customer_experience Tables: kitsune_retrieval_index, zendesk_retrieval_index, knowledge_base_retrieval_index

Date filter columns:

• Kitsune: creation_date DATE — use creation_date BETWEEN 'YYYY-MM-DD' AND 'YYYY-MM-DD'
• Zendesk: creation_date DATE — use creation_date BETWEEN 'YYYY-MM-DD' AND 'YYYY-MM-DD'
• Knowledge Base: no creation_date; use last_approved_revision_date (DATE) if a date bound is needed

-- Top topics by volume (Kitsune)
SELECT topic, COUNT(*) AS count
FROM `mozdata.customer_experience.kitsune_retrieval_index`
WHERE creation_date BETWEEN '2026-03-24' AND '2026-04-22'
GROUP BY topic ORDER BY count DESC LIMIT 10

-- Top categories by volume (Zendesk)
SELECT ticket_category_llm, COUNT(*) AS count
FROM `mozdata.customer_experience.zendesk_retrieval_index`
WHERE creation_date BETWEEN '2026-03-24' AND '2026-04-22'
GROUP BY ticket_category_llm ORDER BY count DESC LIMIT 10

-- Average sentiment by topic (Kitsune only — do not use Zendesk sentiment)
SELECT topic, COUNT(*) AS count, ROUND(AVG(question_sentiment_score), 2) AS avg_sentiment
FROM `mozdata.customer_experience.kitsune_retrieval_index`
WHERE creation_date BETWEEN '2026-03-24' AND '2026-04-22'
GROUP BY topic ORDER BY avg_sentiment ASC LIMIT 10

-- Volume by product (Zendesk)
SELECT product, COUNT(*) AS count
FROM `mozdata.customer_experience.zendesk_retrieval_index`
WHERE creation_date BETWEEN '2026-03-24' AND '2026-04-22'
GROUP BY product ORDER BY count DESC

Output

A plain-text formatted table:

topic          count
-------------  -----
site-breakage  142
sync           98
passwords      74

Troubleshooting