From Claude DB
Audits column types and precision in database schemas: flags float money (severity-5), naive timestamps, jsonb schema evasion, enum vs lookup, and charset/collation issues. Use for schema review.
How this skill is triggered — by the user, by Claude, or both
Slash command
/claude-db:db-types-precisionThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
The type is the first and cheapest constraint: it decides what values are even representable and how they sort, compare, and round. The classic data-loss bug — money in a float — lives here. This module is **design**-axis (Tipos category, shared with M6). It applies across paradigms with paradigm-appropriate type vocabularies (Postgres `numeric`, Mongo `Decimal128`).
The type is the first and cheapest constraint: it decides what values are even representable and how they sort, compare, and round. The classic data-loss bug — money in a float — lives here. This module is design-axis (Tipos category, shared with M6). It applies across paradigms with paradigm-appropriate type vocabularies (Postgres numeric, Mongo Decimal128).
float/real/double precision (or Mongo Double) for currency/amounts — binary floating point cannot represent decimal cents exactly. Use numeric/decimal/Decimal128/integer-minor-units. This is the severity-5 cap.timestamp/datetime (without time zone) for instants that cross zones; prefer timestamptz stored in UTC. Flag timestamp columns named *_at lacking tz.jsonb/json column carrying what should be first-class typed/constrained columns (stable, queried, FK-related keys buried in JSON). Embedding flexible blobs is fine; hiding the schema is not.ENUM (esp. MySQL ENUM, hard to alter) where a referenced lookup table would be safer to evolve; or free-text status where a constrained domain is needed.utf8 (3-byte, no emoji/astral) instead of utf8mb4; an unintended case-sensitive or accent-sensitive collation on identifiers/emails.fail, axis design, confidence established (caps the Design score).warn, axis design.warn, fixable: proposed.utf8 (not utf8mb4): severity 3, warn.warn.Parse DDL/snapshot via scripts/parse-schema.mjs: match money-like column names (price, amount, total, balance, cost, *_cents) against float|real|double|Double types; flag timestamp/datetime without tz on *_at columns; detect json/jsonb columns and ENUM(...); read declared charset/collation. Directional program-source parses never raise the money-float sev-5 cap.
Confirm float money columns from the catalog:
-- $DATABASE_URL, read-only
SELECT table_name, column_name, data_type
FROM information_schema.columns
WHERE data_type IN ('real','double precision')
AND column_name ~* '(price|amount|total|balance|cost|fee|tax)';
MySQL charset check:
SELECT table_name, column_name, character_set_name, collation_name
FROM information_schema.columns WHERE character_set_name = 'utf8';
Emit per schema/finding.schema.json. Examples:
M4.orders.total_float_money — total double precision loses cents (severity 5, fail, axis design, confidence established).M4.events.created_at_naive_timestamp — created_at timestamp lacks time zone (severity 4, warn, axis design, fixable: proposed).M4.products.attributes_jsonb_schema_evasion — stable queried keys buried in jsonb (severity 3, warn).M4.users.email_utf8_not_utf8mb4 — MySQL utf8 truncates astral chars (severity 3, warn).
Each finding: evidence.observed quotes the column DDL verbatim; verification.reproduce is a runnable query above (method: ddl_parse / schema_introspect); expected_impact banded + confidence-tagged.Double); never soften it.jsonb is the right tool for genuinely flexible/sparse data — only flag it when it conceals a stable, queried, relational schema.npx claudepluginhub hainrixz/claude-db --plugin claude-dbProvides UI/UX resources: 50+ styles, color palettes, font pairings, guidelines, charts for web/mobile across React, Next.js, Vue, Svelte, Tailwind, React Native, Flutter. Aids planning, building, reviewing interfaces.
Searches MemPalace before answering questions about past work, people, projects, or prior decisions. Returns verbatim stored content instead of guessing from model memory.