skillshop-ostyles/agents-data

Schema-vs-usage contract auditor: compares schema declarations (DDL, ORM models, TypeScript interfaces, GraphQL types, OpenAPI schemas) against actual usage sites (API responses, form handlers, import parsers, ORM writes) and has the LLM judge whether each violation is a real contract break, harmless flexibility, or schema too strict. Read-only. Trigger: /data-contract

Data fixture auditor: scans test fixture files for coverage gaps, stale data shapes, and missing edge cases. Audits JSON, YAML, and SQL fixtures for cardinality, null ratio, and schema drift. Read-only. Audience: Senior. Trigger: /fixture-audit

Database migration safety inspector: scans SQL migration files for 20+ safety rules (table rewrites, missing CONCURRENTLY, destructive DROP, unsafe constraints, missing IF EXISTS, DML on existing tables, missing transaction wrappers), then the LLM assesses blast radius per finding against actual schema and usage patterns, recommending safer alternatives. Read-only. Trigger: /migration-safety

Schema migration surgeon: diffs two schema states (SQL DDL or Prisma), then generates the complete package nobody writes by hand - forward migration, rollback, pre/post validation queries and a risk protocol with explicit data-loss warnings. NEVER executes anything against a database; generates files only. Trigger: /migrate

Migration test writer: reads a schema diff (old DDL vs. new DDL), identifies structural changes, and generates pre- and post-migration validation queries. Read-only. Audience: Senior. Trigger: /migration-test

N+1 query hunter: traces loop-to-query data flows in source code, identifies ORM N+1 patterns, and distinguishes real N+1 problems from intentional batch/join patterns. Read-only. Audience: Both. Trigger: /n-plus-one

PII schema classifier: scans DDL/ORM models for columns that may contain sensitive data, classifies by sensitivity level using naming patterns, and flags ambiguous columns for LLM domain review. Read-only. Audience: Senior. Trigger: /pii-scan

Relationship inference: scans DDL and code for missing foreign key relationships, infers relationships from naming patterns and query join patterns, and validates each inference against business logic. Read-only. Audience: Senior. Trigger: /infer-rels

Generate human-readable data dictionary from DDL with LLM-written business descriptions. Reads DDL or ORM models, extracts structural metadata, and writes plain-English descriptions for every table, column, and relationship. Trigger: /schema-docs

Schema drift is the silent erosion of database integrity: a column added to production but not to staging, a NOT NULL that became nullable via a hotfix, a dev-only index that made it to production. This skill compares two schema snapshots (DDL) and the LLM assesses the criticality of each drift.

Database schema health scanner: parses DDL files (.sql, .prisma, ORM models), extracts per-table structural metrics (columns, PKs, FKs, indexes, naming conventions, type consistency), detects anti-patterns (missing PKs, unbounded strings, mixed naming, god tables, missing timestamps, FK without index), then the LLM judges each finding in domain context. Read-only. Audience: Both. Trigger: /schema-health

Seed data strategist: reads a schema (DDL/ORM models), analyzes tables, columns, constraints, and enum values, and generates a comprehensive seed data strategy with meaningful test scenarios. Read-only. Audience: Both. Trigger: /seed-data

Inline SQL smell detector: harvests SQL strings from application code, runs 15+ static analysis rules (SELECT *, missing WHERE, implicit casts, non-sargable filters, cartesian products, SELECT DISTINCT masking bad joins), then LLM classifies business impact and proposes rewritten SQL. Read-only. Trigger: /sql-smells

Semantic test gap mapper: inventories the public code surface (exports, routes) and all existing tests, then has the LLM map which BEHAVIORS of each public symbol are covered by which test and which are not - reporting untested behaviors (edge cases, error paths, boundaries) ranked by risk, with proposed test case names. Static, never runs tests. Read-only. Trigger: /testgap