clickhouse-architecture-advisor Skill for Claude: What It Does and How to Install It
Makes Claude design ClickHouse architectures for a workload, and label every recommendation as official, derived or field advice.
- Installs
- 10.2K
- on skills.sh
- Stars
- 565
- on GitHub
- Files
- 14
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
This skill is ClickHouse's own architecture advisor. Claude first names the workload shape (observability, SIEM, product analytics, IoT, market data or mixed), reads the matching decision rule files and attaches official doc links.
Every recommendation is tagged official, derived or field, with a confidence level, and field advice is flagged as heuristic. The skill is meant to sit on top of the separate clickhouse-best-practices skill. It bundles no scripts.
Good for
- Engineers designing real-time analytics on ClickHouse
- Architects running a ClickHouse workshop or proof of concept
- Teams choosing between async inserts, Kafka and pre-aggregation
Install clickhouse-architecture-advisor
Claude Code
- 1
Copy this prompt.
Install the clickhouse-architecture-advisor skill in Claude Code. Run this command: /plugin marketplace add clickhouse/agent-skills /plugin install clickhouse-best-practices@clickhouse-agent-skills Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Pick one route. SKILL.md says this skill adds architecture decisions on top of clickhouse-best-practices and tells Claude to read that skill's rules, so the routes here install both. The clickhouse-best-practices plugin's plugin.json lists 8 skills, including both of these. The architecture-advisor plugin installs this skill alone. The skill bundles no scripts.
Built from the package data we hold. Not yet tested by us. Source.
What's in the skill
When it loads
MUST USE when designing ClickHouse architectures, selecting between ingestion or modeling patterns, or translating best practices into workload-specific system designs. Complements clickhouse-best-practices with decision frameworks and explicit provenance labels.
Sections in SKILL.md
- Required behavior
- Provenance rules
- official
- derived
- field
- Read these rule files by scenario
- Real-time ingestion design
- Time-series and retention design
- Enrichment and dimension lookups
- Mutable state / late-arriving events
- Output format
- Workload Summary
- Key Decisions
- Recommendations
- <Recommendation title>
- Architecture-specific guidance
Key facts
- Installs on skills.sh
- 10,223 (Nov 2, 2026)
- Last repo commit
- Oct 30, 2026
What it includes
Workload shape, rule files, doc links, provenance labels
Six steps: identify one of 6 workload shapes, read the rules, attach docs, label each recommendation, never present field advice as official, and say when unsure.
Try: “Design ClickHouse ingestion and storage for 200K log events per second.”
Read ·
Required behaviorDirect inserts, async inserts, Kafka engine with a materialized view
decision-ingestion-strategy.md compares paths by throughput, latency and producer shape.
Read ·
Ingestion strategyRaw only, incremental or refreshable materialized views, rollups
decision-real-time-preaggregation.md.
Read ·
Real-time pre-aggregationWhether to partition, granularity, TTL and partition counts
decision-partitioning-timeseries.md.
Read ·
Time-series partitioningRuntime join, dictionary, denormalization or materialized
decision-join-enrichment.md.
Read ·
Enrichment and joinsAppend-only events, replacing or collapsing semantics
decision-late-arriving-upserts.md.
Read ·
Late-arriving data and upsertsWorkload summary, key decisions, recommendations with source and validation
Each recommendation has what, why, how, category, confidence, source and a validation step such as SQL or a metric.
Read ·
Output formatObservability, financial market surveillance and SIEM
Three sample outputs in examples/.
Read ·
Example answers
From the official docs
Questions
What does the clickhouse-architecture-advisor skill do?
It makes Claude plan a ClickHouse design for your workload. Claude weighs ingestion, partitioning, joins, upserts and pre-aggregation, and labels each recommendation as official, derived or field advice.
How do I install it?
Use the install box at the top of this page. It installs this skill with clickhouse-best-practices, which SKILL.md says it builds on.
Is this skill from ClickHouse?
Yes. It is in ClickHouse's own repo, clickhouse/agent-skills, and SKILL.md names ClickHouse Inc as author.
What do official, derived and field mean?
Official is backed directly by ClickHouse docs. Derived follows logically from documented behavior. Field is experience-based and must carry a disclaimer that it is heuristic.
What is the difference from clickhouse-best-practices?
The best-practices skill checks concrete schema and query rules. This one decides when, why and how: the overall architecture and its tradeoffs. The README says it complements, not replaces, that skill.
Does it work in the Claude app?
Yes. The skill bundles no scripts, so the install box has a Claude app upload for it and the best-practices skill.
Is it safe?
It bundles no scripts. The workload details you describe go to your assistant as part of the conversation, so leave out hostnames, credentials and customer data. Its recommendations are advice for you to check: each one carries a validation step, such as a SQL query or a metric, and field advice is flagged as heuristic.
Information
updated Nov 2, 2026
- Repository
- clickhouse/agent-skills
- Folder
skills/clickhouse-architecture-advisor
Sources
Docs, issue trackers and discussions this page is built from.
Show all 5Hide
- clickhouse-architecture-advisor SKILL.md · github.com
- clickhouse/agent-skills README · github.com
- marketplace.json · github.com
- clickhouse-best-practices plugin.json · github.com
- advisor README · github.com