What problem does it solve? When a spec, premix, cheese, or shipping Excel import produces wrong, mislinked, misnamed, duplicated, or skipped data, it is hard to know which layer failed. This Skill provides a standard drill to isolate whether the fault lies in the AI parse, the apply/tie logic, the pool, or the profile before any code is changed. ## Core Features & Use Cases - Layered Diagnosis: Inspect saved parse sheets (saved_spec_sheets, saved_premix_sheets, saved_shipping_guides) via SQL, then compare against brand_profiles, recipe pools, and learned aliases to classify the failure layer. - Targeted Fix Routing: Route parse errors to lib/spec-import prompt/pipeline fixes (with SPEC_PARSE_VERSION bumps), apply/tie bugs to applySpecImport in storage.ts, and pool errors to link/dedupe passes. - Sharp-Edge Awareness: Encodes domain rules such as dough family collapse, blank-fill-only relinked ties, import order and dedup keys, and alias priority (alias > AI > fuzzy). - Use Case: A user reports that a dough spec import created duplicate recipes with wrong batch yields. The Skill guides checking the saved parse JSON first, confirming the parse was correct, then tracing the bug to the tie logic and following the data-heal-playbook for already-landed bad data. ## Quick Start Ask the assistant to investigate why the latest cheese spec import linked the wrong brand profile, and have it trace the saved parse against what landed in the database.
