Convert LookML views, models, and explores into ktx semantic layer sources.
npx skills add https://github.com/Kaelio/ktx --skill lookml-ingest
Please help me install this Agent Skill. Skill: lookml_ingest Source: https://github.com/Kaelio/ktx/tree/main/packages/cli/src/skills/lookml_ingest Command: npx skills add https://github.com/Kaelio/ktx --skill lookml-ingest
Translating LookML definitions into a semantic layer by hand is error-prone: dimension_group timeframes get mistaken for physical columns, derived tables lose their grain, and enforced filters like sql_always_where get dropped. This Skill provides a deterministic mapping from LookML constructs to ktx semantic layer sources with verification steps that prevent hallucinated column names.
Load this skill whenever the turn contains .lkml content and ask the agent to map the LookML views and explores into ktx semantic layer sources with provenance tags.
High-intent search queries and answers about installing and using this skill.
Map each LookML view to either an overlay (when sql_table_name matches a manifest table) or a standalone source (when it uses derived_table or sql_always_where). Measures become measures, explore joins become join entries, and always_filter or conditionally_filter become named segments.
A dimension_group expands at Looker render time into fields like date_raw or date_week, but these are not physical warehouse columns. Map each dimension_group to a single column entry using the physical column from its sql clause, with type time and role time.
Use an overlay when the view is a thin wrapper over a manifest-backed table via sql_table_name, inheriting grain and base columns. Use a standalone source with explicit sql, grain, and columns when the view defines a derived_table or sql_always_where.
These errors occur when invented column names, such as timeframe expansions like date_date, are written into sql, columns, or grain. Prevent this by calling sl_discover or a sql_execution probe before writing, and sl_validate after every sl_write_source.
A hard gate marked [LOOKML SL WRITES DISALLOWED] blocks sl_write_source and sl_edit_source for that WorkUnit when the declared Looker connection mismatches the expected connection name. Continue wiki extraction and preserve the mismatch reason in any emit_unmapped_fallback.
Fold sql_always_where directly into the standalone source's sql clause as a WHERE condition, since it is enforced rather than opt-in. Do not model it as a segment or duplicate it as per-measure filters.
PostHogConnect external data warehouse sources to PostHog with per-table sync methods.
kyle-chalmersAnswer business analytics questions by querying dbt's Semantic Layer or ad-hoc SQL.
malloydataConvert LookML models to Malloy semantic models by analyzing business logic and relationships.
sigmacomputingConvert Looker LookML models and dashboards into Sigma data models and workbooks.
