What problem does it solve? Query results from a data warehouse can be silently wrong due to bad joins, fanout, overly restrictive filters, or misunderstood business rules, and hard-won context about tables and fields is lost after each session. ## Core Features & Use Cases - Plausibility Checks: Automatically inspects every execute_sql result for row count anomalies, invalid value ranges, out-of-bounds dates, grain mismatches, magnitude drift versus past queries, and high null rates. - Validation Rules: Enforces custom validation rules defined on field files and lore files in the data model vault, catching domain-specific mistakes that generic checks miss. - Self-Learning Proposals: Maintains a scratchpad of unmapped columns, guessed joins, and contradictions, then batches proposals for new or updated vault files after presenting results. - Use Case: After running a monthly revenue query, the skill flags that revenue contains unexpected negative values, retries the query with a corrected join, and then proposes a new relation file documenting the join for future queries. ## Quick Start Ask the agent to check the results of your last SQL query for plausibility issues and propose any missing data model files.
