Extracts ktx knowledge and semantic-layer proposals from staged Looker dashboard, Look, and explore JSON.
npx skills add https://github.com/Kaelio/ktx --skill looker-ingest
Please help me install this Agent Skill. Skill: looker_ingest Source: https://github.com/Kaelio/ktx/tree/main/packages/cli/src/skills/looker_ingest Command: npx skills add https://github.com/Kaelio/ktx --skill looker-ingest
Looker instances accumulate years of metric definitions, segment logic, and business semantics locked inside dashboards, Looks, and explores. This Skill converts that staged Looker runtime JSON into durable ktx wiki knowledge and semantic-layer sources, so agents can query the warehouse using the same definitions your BI teams already trust.
looker__<model>__<explore> key, branching on the staged targetTable contract and validating every write with sl_validate.looker_query_to_sl on each Look or dashboard tile query, then adds measures, creates sources, or falls back to wiki-only candidates based on the proposal decision.schema.table.column reference with entity_details and sql_execution probes before promoting it into wiki pages or SL sources, preventing fictional identifiers.explores/, dashboards/, and looks/, run this Skill to turn a high-usage "Open Pipeline ARR" Look into a validated SL measure plus a wiki definition, while skipping low-usage ad-hoc views.Load the looker_ingest skill and process the staged Looker WorkUnit whose raw files are under explores/, dashboards/, or looks/ to produce wiki knowledge candidates and validated semantic-layer sources.
High-intent search queries and answers about installing and using this skill.
Stage the explore JSON under explores/<model>/<explore>.json, then map it to a source named looker__<model>__<explore>. When targetTable.ok is true, write the source on targetWarehouseConnectionId using targetTable.canonicalTable as source.table, and run sl_validate afterward.
Read every rawFiles entry for the WorkUnit, call looker_query_to_sl for each query object, and branch on the proposal decision. Write measures or sources only when canWriteStandaloneSource is true, otherwise record wiki candidates or emit_unmapped_fallback.
The WorkUnit stays wiki-only for SL purposes. Capture durable domain semantics with context_candidate_write, then emit_unmapped_fallback with the exact reason code such as no_connection_mapping, putting human-readable context in clarification.
No. Usage counts, owner names, schedules, and favorites are used only for prioritization and candidate scoring. They must never appear in assertion, rationale, or wiki prose; attribution stays in the provenance action trail.
Raw Looker sql_table_name values can contain aliases, ${TABLE} templates, or derived-table SQL that do not resolve as physical tables. Verify with entity_details and a sql_execution probe; if unresolved, mark the identifier unverified or emit a fallback with the failing probe error.
Skip point-in-time values, dashboard layout and visualization settings, ad-hoc low-usage queries, and simple saved views with no metric, segment, or concept definition. Zero or near-zero usage lowers priority and usually means skip.
abhiroopbQuery and manage Looker dashboards, tiles, queries, and explores via Python CLI.
AradotsoBuild a semantic context layer for AI agents querying data warehouses.
sigmacomputingConvert Looker LookML models and dashboards into Sigma data models and workbooks.
sparkline-venturesValidate product-eval scoring constants against known decision outcomes.