Agent Skills: Relational Database Design

Relational Database Design

A logical, RDBMS-agnostic method for designing sound relational databases — the decisions a schema forces and how a disciplined logical-design methodology settles them, written so an agent applies them while designing or reviewing a schema. Each rule names a specific wrong default it corrects; there is no rule for things the model already gets right.

The whole method produces the logical structure first — the tables, fields, keys, relationships, and integrity rules an organization's information requires — deliberately independent of any particular RDBMS product or physical/performance concern. Followed faithfully, it yields fully normalized tables without treating normalization as a separate back-end phase.

When to Apply

• Designing a new relational schema from requirements
• Reviewing or refactoring an existing schema for structural soundness
• Resolving repeating groups, multivalued/multipart fields, or redundant data
• Choosing candidate, primary, and foreign keys for a table
• Modeling one-to-one, one-to-many, many-to-many, or self-referencing relationships
• Deciding deletion rules (restrict, cascade, nullify, deny, set default) and participation constraints
• Deciding where a constraint belongs — field spec, relationship, validation table, or application
• Diagnosing a schema that duplicates, loses, or corrupts data

This skill covers logical design. It does not cover SQL dialects, indexing, partitioning, query tuning, or analytical/dimensional (star-schema) modeling — those are physical/implementation concerns handled after the logical design is sound.

Rule Categories

| # | Category | Prefix | Covers | |---|----------|--------|--------| | 1 | Design Process & Requirements | proc- | Design logically before choosing an RDBMS, follow the sequence, drive from mission + analysis, normalization is built in | | 2 | Table Structure | tbl- | One subject per table, the ideal table, no reference fields, minimal redundancy | | 3 | Field Design | fld- | The ideal field, single values, atomic fields, no stored calculations, clear names | | 4 | Keys | key- | Candidate key elements, one primary key per table, foreign keys mirror primary keys, stable non-sensitive keys | | 5 | Relationships | rel- | Junction tables for many-to-many, foreign-key placement, deletion rules, participation, self-referencing | | 6 | Data Integrity, Rules & Views | intg- | The four integrity levels, field specifications, database-vs-application rules, validation tables, views | | 7 | Antipatterns | anti- | Flat-file, spreadsheet-as-database, RDBMS-driven design, when bending the rules is defensible | | 8 | Terminology | term- | Data vs information, nulls, core relational vocabulary |

Quick Reference

1. Design Process & Requirements

• proc-follow-the-design-sequence — later steps depend on earlier ones; skipping steps yields poor integrity
• proc-design-logically-before-choosing-rdbms — design the logical structure with no RDBMS in mind; choose and implement afterward
• proc-start-from-mission-and-objectives — a mission statement and objectives scope the database and reveal its subjects
• proc-analyze-and-interview-before-designing — analyze the current system and interview users and management before inventing fields
• proc-normalization-is-built-in — the ideal-field/ideal-table/key guidelines yield normalized tables; normalization is not a separate phase

2. Table Structure

• tbl-one-subject-per-table — each table represents exactly one subject (object or event)
• tbl-ideal-table-checklist — the six-point test for a sound table structure
• tbl-no-reference-fields — don't copy fields from another table for reporting convenience
• tbl-minimize-redundant-data — FK-driven redundancy is fine; keep everything else to an absolute minimum

3. Field Design

• fld-ideal-field-checklist — the test for a sound field
• fld-store-single-values-not-lists — a multivalued field becomes its own table
• fld-keep-fields-atomic — decompose multipart/composite fields into one field per distinct item
• fld-derive-dont-store-calculated-values — calculations belong in a view, not a stored field
• fld-clear-singular-field-names — one unambiguous, singular name per characteristic

4. Keys

• key-candidate-key-elements — the test a field (or field set) must pass to be a candidate key
• key-one-primary-key-per-table — exactly one non-null, unique, stable primary key per table
• key-foreign-key-mirrors-its-primary-key — same name, replica spec, values drawn from the referenced primary key
• key-prefer-stable-non-sensitive-keys — a key value should rarely change and must not expose sensitive data

5. Relationships

• rel-junction-table-for-many-to-many — resolve M:N with a linking table keyed on both foreign keys
• rel-foreign-key-on-the-many-side — the foreign key lives on the many side of a one-to-many
• rel-deletion-rule-guards-orphans — every relationship gets a deletion rule; restrict by default
• rel-participation-encodes-constraints — mandatory/optional and degree (min,max) capture real business limits
• rel-self-referencing-relationships — a table related to itself needs a distinctly named foreign key

6. Data Integrity, Rules & Views

• intg-four-levels-of-data-integrity — table, field, relationship, and business-rule integrity together
• intg-field-specifications — pin down every field's general, physical, and logical elements
• intg-database-vs-application-rules — enforce structural rules in the schema; conditional/derived rules in the application
• intg-validation-tables-for-allowed-values — a lookup table beats a hardcoded value list
• intg-views-for-derived-and-restricted-data — data, aggregate, and validation views instead of extra stored fields

7. Antipatterns

• anti-flat-file-design — the throw-everything-in-one-table structure and how to spot it
• anti-spreadsheet-as-database — a spreadsheet layout is not a relational schema
• anti-rdbms-driven-design — don't let the product you know dictate the design
• anti-bend-rules-only-deliberately — the only two defensible reasons to break the rules, and how to document it

8. Terminology

• term-data-versus-information — data is stored; information is data made meaningful for presentation
• term-null-is-unknown-not-zero — a null is a missing/unknown value, not zero or blank
• term-relational-vocabulary — key is logical, index is physical; a view is not a table

How to Use

Read a reference file when its decision comes up. Each rule names the wrong default it corrects, then shows the canonical way (with an incorrect/correct contrast only where the wrong way is a real trap).

• Section definitions — category structure
• Rule template — for adding new rules
• AGENTS.md — auto-built table of contents across all rules

Reference Files

| File | Description | |------|-------------| | references/_sections.md | Category definitions and ordering | | assets/templates/_template.md | Template for new rules | | metadata.json | Version and source references |