clickhouse/agent-skills
5 skills7.2k installs2.3k starsGitHub
Install
npx skills add https://github.com/clickhouse/agent-skillsSkills in this repo
1Clickhouse Best PracticesClickHouse Best Practices is an agent-oriented reference for solo builders and small teams running analytics on ClickHouse 24.1+. It packages schema design, query tuning, table engines, indexing, materialized views, and cluster operations into explicit rules, each with SQL contrasts and stated impact so an agent does not guess at engine or PARTITION choices. Use it when you are creating new fact tables, fixing slow scans, or refactoring Nullable columns and ORDER BY keys before scale bites. It fits SaaS and API products that ship event or metrics stores, and CLI or internal tools that embed ClickHouse. The skill is not a managed connector; it is procedural knowledge meant to keep automated migrations and review loops consistent with ClickHouse Inc guidance from early 2026.3.5kinstalls2Clickhouse Architecture AdvisorClickHouse Architecture Advisor is an agent skill for solo and indie builders who need structured architecture guidance beyond a static best-practices checklist. It is built for advisory sessions, workshops, and early system design when you must choose ingestion paths (direct inserts, async inserts, Kafka engine plus materialized views, or upstream buffering), time-series partitioning and TTL, enrichment (runtime JOINs, dictionaries, denormalization, or materialized enrichment), late-arriving and mutable data semantics, and real-time pre-aggregation (incremental MVs, refreshable MVs, rollups). Each answer follows an output standard and labels guidance as official, derived, or field so you can cite defensible rationale in plans and PRs. Use it while scoping analytics backends, implementing pipelines, or revisiting production schemas when workload shape changes.1.2kinstalls3Clickhousectl Local Devclickhousectl-local-dev is an agent skill published by ClickHouse Inc that walks solo builders through a complete local ClickHouse development environment using the clickhousectl CLI. It is for anyone building SaaS, APIs, or data-heavy agents who need columnar analytics on their machine before cloud spend. Invoke when the user asks to install ClickHouse, create a local server, define tables, seed data, or experiment with ClickHouse for the first time. The skill enforces a sequential checklist—verify the CLI exists, install components, initialize the project, bring the server up, then validate queries—so agents do not skip foundational steps that break later integrations. That matters for indie teams who lack a dedicated data engineer and rely on the coding agent to mirror official tooling. Apache-2.0 licensed and aligned with ClickHouse documentation, it reduces wrong-connection strings and half-configured daemons that waste debug time during Build.867installs4Chdb Sqlchdb SQL is an agent skill packaged as runnable cookbook examples for the chdb embedded ClickHouse engine in Python. Solo builders use it to query arbitrary files—Parquet, CSV, JSON Lines—and combine them with SQL joins, window functions, and UDFs without provisioning a database cluster. The readme walks through parametrized queries, registering Python data as tables, building analytical tables in a session, and streaming when result sets exceed memory. It is ideal when you are validating analytics on exports, log shards, or product telemetry dumps during development, and when you want copy-paste patterns that show expected output in comments. Grow-phase reporting and Operate-phase log triage both benefit once the same SQL migrates to ClickHouse Cloud or self-hosted CH, but the canonical shelf remains backend analytics prototyping.809installs5Chdb Datastorechdb-datastore is an agent skill that documents how to use ClickHouse’s embedded chdb DataStore as a pandas-accelerated analytics layer for solo and indie builders shipping data-heavy features. Instead of standing up a full warehouse before you validate a metric, you change a single import and run SQL-style joins across local Parquet, operational MySQL or PostgreSQL, and cloud buckets from one Python surface. The readme ships eleven self-contained example sections—from a minimal pandas replacement through cross-source writes and remote schema exploration—so coding agents can copy patterns with expected output comments. It matters when you need credible dashboards, internal ops reports, or agent tools that fuse product DB rows with files in S3 without a big infra project. Complexity sits at intermediate: you should be comfortable with Python, basic SQL, and credentials for the systems you join.774installs