What problem does it solve? Reviewing Malloy semantic models by hand is slow and error-prone: silent cardinality bugs, non-unique primary keys, broken chart tags, and missing documentation all compile cleanly but produce wrong numbers or broken dashboards. This Skill automates a single-pass, rubric-driven audit of .malloy files and produces a navigable Markdown review file. ## Core Features & Use Cases - Rubric-based findings: Applies seven dimension rubrics (correctness, documentation, style, structure, queries, rendering, governance) with severity, confidence scoring, and suggested fixes, dropping findings below 80% confidence. - Data-verified checks: Uses execute_query to verify declared primary_key: uniqueness and search_malloy_docs to validate Malloy capabilities before flagging raw SQL. - Multiple review modes: Single-file mode, directory audit mode, and GitHub PR mode (--pr <n>) with optional --comment posting via gh pr comment. - Use Case: A data engineer opens a PR touching 14 Malloy model files. Run the review in PR mode to get a triage document with blockers, cross-cutting themes, per-file findings, and a machine-readable JSON tail. ## Quick Start Ask the assistant to review a Malloy file or folder, for example: run a Malloy code review on the models/finance directory and write the findings to a review file.
