ktx Analytics Workflow

What it does

It gives Claude a standard operating procedure for answering any data question through the ktx MCP tools (discover_data, entity_details, dictionary_search, sl_query, sql_execution, memory_ingest).

• Enforces a 7-step loop — discover → inspect → resolve values → plan → query → validate → capture learnings — so table and column names are never guessed.
• Prefers the semantic layer (measures/dimensions) over raw SQL, keeping metric definitions authoritative.
• The <sql_craft> section is a long, example-backed catalogue of the errors that produce silently wrong numbers: fan-out joins inflating SUMs, inner-joining a dimension purely for a label and dropping fact rows, parsing text-encoded numerics (1.2K, $1,200, -), the RANGE-frame trap in running totals, integer division on postgres/sqlite/tsql, filtering before a window, macro vs micro averages, and period anchoring on lifecycle timestamps.
• Requires the exact output-column contract to be written down before the SQL, plus a final completeness check before answering.

Who it's for

• Teams already running the ktx CLI/MCP against a warehouse (BigQuery, Postgres, Snowflake, SQLite, ...).
• Analysts and analytics engineers who let an LLM write ad-hoc SQL and have been burned by subtly wrong totals.
• Data orgs that want metric definitions and schema gotchas accumulated back into a searchable memory/wiki.

Examples

1. "How many orders per region contain a returned item?" — the skill pre-aggregates order lines to one row per qualifying order before joining, so an order with two returned lines isn't counted twice.
2. "Each customer's first order, but only since 2024-01-01" — the window is computed over the full partition and filtered in the outer query, so "first" reflects real history.
3. "What's total trade volume?" when amounts are stored as '3M' / '$1,200' / '-' — one early CTE strips symbols, scales K/M/B suffixes and maps sentinels, avoiding a SUM that silently collapses to near-zero.

Install the skill I found on Claude Skill Mart.
Copy the packages/cli/src/skills/analytics folder from the GitHub repo Kaelio/ktx into my ~/.claude/skills/ktx-analytics/.
When it's done, tell me in one line what this skill can do.