ClickHouse Analyst (Sub-Agent Architecture)

This skill uses focused agents (often run in waves) to diagnose ClickHouse issues quickly. Each agent:

1. Runs SQL queries directly against ClickHouse (via a backend: CLI or MCP)
2. Analyzes results with an LLM
3. Produces reproducible artifacts and an RCA-style report

How to Use

Pick an execution backend (do this first)

Before running any agents, decide how queries will be executed in this environment. Then load exactly one backend doc and follow it for execution details.

Choose a backend using these rules:

• BACKEND-MCP.md use when you cannot spawn/exec processes and cannot run clickhouse-client, but you do have an MCP ClickHouse connector tool available (WebUI-style environments).
• BACKEND-CLI.md use when you can spawn/exec locally and have clickhouse-client available (terminal/SSH environments).

Prefer BACKEND-CLI when a spawn/exec tool is available. Prefer BACKEND-MCP when process spawning is restricted or as a fallback if clickhouse-client is not available or has connectivity problems.