What problem does it solve? Answering data questions against Malloy semantic models without direct database access is error-prone: guessed field names, fan-out double-counting, broken filters, and unverified results produce plausible but wrong answers. This Skill enforces a disciplined analyst workflow so every answer is grounded in the actual model and verified before it is presented. ## Core Features & Use Cases - Structured discovery: Uses get_context to confirm entities, read source-level and field-level docstrings for grain, units, and null handling, and never guesses names. - Query construction and execution: Builds Malloy queries with execute_query, avoiding common compile errors like count(distinct), semicolon separators, and invalid join paths. - Verification before delivery: Cross-checks results with independent recounts, denominator checks, fan-out probes, and date-range grounding before presenting. - Use Case: Ask "how many active subscribers did we have last month?" and the Skill discovers the right measure and dimensions, runs a validated Malloy query, verifies the count independently, and presents the answer with stated assumptions. ## Quick Start Ask a natural-language data question about a published Credible workspace, such as "break down revenue by region for Q1", and let the Skill discover the model, run, and verify the query.
