Answers data questions by querying ktx-connected warehouses through semantic-layer tools and read-only SQL.
npx skills add https://github.com/Kaelio/ktx --skill ktx-analytics
Please help me install this Agent Skill. Skill: ktx-analytics Source: https://github.com/Kaelio/ktx/tree/main/packages/cli/src/skills/analytics Command: npx skills add https://github.com/Kaelio/ktx --skill ktx-analytics
Analysts and agents often guess table names, misread column encodings, or write SQL that runs but returns silently wrong numbers. This Skill provides a disciplined workflow for answering data questions against ktx-connected databases, using governed metric definitions and validated SQL instead of guesswork.
discover_data first to locate wiki pages, semantic-layer sources, metrics, tables, and columns before writing any SQL.sl_query over raw SQL so approved measures remain the source of truth, falling back to read-only sql_execution only when needed.dictionary_search, queries the semantic layer, validates the result, and captures durable learnings via memory_ingest.Ask a data question such as "show me monthly active users by plan tier for the last six months" against your configured ktx connection.
High-intent search queries and answers about installing and using this skill.
Start by calling discover_data to see available wiki pages, semantic-layer sources, metrics, tables, and columns. Then inspect promising refs, resolve business values with dictionary_search, and query with sl_query or read-only sql_execution.
Prefer sl_query whenever the semantic layer covers the question, since approved measures are the source of truth. Use sql_execution only for questions the semantic layer does not cover, after fetching engine conventions with sql_dialect_notes.
No. The sql_execution tool is strictly read-only and the server rejects write statements. The Skill is designed for analysis, exploration, and metric investigation only.
Common causes include fan-out joins inflating sums, integer division truncating rates, text-encoded numerics sorting lexically, and filtering before window functions. The Skill's sql_craft rules address each of these with sampling, pre-aggregation, casting, and explicit window frames.
Pass connectionId to entity_details, sl_read_source, and sql_execution when intent pins a specific warehouse. Omit it for unscoped discovery calls like discover_data and dictionary_search, and ask the user which warehouse to use when scoping is ambiguous.
AradotsoBuild a semantic context layer for AI agents querying data warehouses.
lightdashQuery governed metrics and create charts through the Lightdash MCP server.
xpert-aiGather business context from sources to frame downstream data analysis.
streamlitCreate Streamlit dashboards with reusable card-based layouts and KPI components.