Rust Persistence, SQLx, And SeaQuery

Use this skill for database-backed Rust work. Keep persistence adapters explicit, keep core domain rules independent of database mechanics where practical, and validate both Rust types and database behavior. For database-neutral SQL design, use sql-engineering. For language-independent PostgreSQL schema, SQL, query-plan, security, and migration design, use postgresql-sql-engineering. For MySQL- or MariaDB-native schema, query-plan, and migration design, use mysql-mariadb-sql-engineering. For SQLite-native schema, transaction, and test-database behavior, use sqlite-sql-engineering. Use hexagonal-architecture when deciding whether repositories, database gateways, transactions, or query services should be outbound ports and adapters rather than core domain code.

For a requested Rust review, use code-review with rust-code-review and review-verification-protocol as the review workflow; add this skill for persistence-specific evidence.

Workflow

1. Inspect the persistence setup: Cargo.toml SQLx and SeaQuery dependencies, feature flags, migrations, DATABASE_URL conventions, .sqlx/ offline metadata, pool types, transaction helpers, repository recipes, seed/fixture strategy, and CI database services.
2. Model the boundary. Use DDD language for tables, aggregates, invariants, and repository/service responsibilities. Do not let query row shapes become the core domain model by accident.
3. When storage behavior changes, design SQL and migrations with the database skill first, then implement the Rust adapter.
4. Implement SQLx or SeaQuery access with typed parameters, explicit result mapping, narrow transactions, and safe error conversion.
5. Test at the right layer: pure domain tests, SQLx integration tests against the target database, and end-to-end tests only for user-visible workflows.
6. Update SQLx offline query metadata for every supported configuration that compiles query macros, or document why offline metadata is not used.

Security Review Prompts

Load security-review when Rust persistence work touches SQL injection risk, SQLx/SeaQuery dynamic SQL, database credentials or DATABASE_URL, migration/backfill exposure, RLS/privilege assumptions, tenant isolation, audit logs, startup migrations, database error leakage, or production-data access. Use threat-modeling before or during new tenant, privilege, migration/backfill, external-database, audit, or sensitive data-flow boundaries. Use dependency-supply-chain-review when persistence crates, migration tools, generated database clients, CI database bootstrap, or database/client binaries raise provenance or advisory questions. Pair security-sensitive reviews with security-review-evidence when evidence includes sanitized SQL, bind values, query plans, .sqlx metadata, migration output, logs, dumps, or other persistence artifacts.

SQLx Checklist

• Prefer compile-checked SQLx macros such as query! or query_as! when the query shape is static and the repository supports online or offline checking.
• Use dynamic SQL only when the shape is genuinely dynamic; bind values instead of formatting user input into SQL.
• Map nullable columns, database enums, time types, UUIDs, JSON, and numeric precision deliberately.
• Keep pool acquisition, transaction scope, and timeout behavior explicit.
• Convert database errors at the adapter edge. Preserve enough context for logs while returning safe domain or API errors.
• For PostgreSQL, MySQL/MariaDB, and SQLite support in one codebase, test each target engine independently. SQL dialects, type affinity, locking, migrations, and constraint behavior differ, and MySQL and MariaDB diverge from each other too even though both use the mysql SQLx/SeaQuery driver.

Offline Metadata

Query macros behind cfg(test), target-specific modules, or features are not necessarily compiled by a default build. Inspect the repository-supported Cargo target and feature matrix, then forward that matrix to Cargo after -- when preparing or checking metadata. --workspace on cargo sqlx prepare writes one workspace-root .sqlx directory; pass Cargo's workspace scope after -- too when the supported command needs it.

For a project where all targets and all features are compatible, for example:

cargo sqlx prepare -- --all-targets --all-features

For a workspace or a narrower supported feature set, adapt the forwarded Cargo arguments rather than assuming that all features can coexist:

cargo sqlx prepare --workspace -- --workspace --all-targets --features <supported-feature-set>
cargo sqlx prepare --check --workspace -- --workspace --all-targets --features <supported-feature-set>
SQLX_OFFLINE=true cargo check --workspace --all-targets --features <supported-feature-set>

Require SQLX_OFFLINE=true compile checks for each supported configuration that uses query macros. Run the matching prepare command against the repository's approved database/schema first, then use prepare --check and the offline Cargo check to prove committed metadata covers that configuration.

Prefer sqlx when:

• Queries are mostly static and readable as SQL.
• Compile-time checked queries or offline query checking are valuable.
• Query shape is known ahead of time.
• Migrations, pools, transactions, and row mapping are the main concern.
• A query builder would obscure simpler explicit SQL.

SeaQuery Checklist

Use SeaQuery for dynamic SQL construction in Rust when it is already a project convention or when runtime query shape would otherwise require unsafe or unreviewable string assembly.

• Inspect the installed version and features before editing: sea-query, sea-query-sqlx, backend features such as backend-postgres, backend-sqlite, or backend-mysql, and existing Iden/identifier enums.
• Prefer SeaQuery when many filters, sorts, joins, projections, predicates, or dialect targets are optional and composition improves clarity.
• Avoid SeaQuery when a plain sqlx::query!, query_as!, or query_file! is clearer, compile-time checking matters more, or dialect-specific SQL is easier to review directly.
• Keep query builders small and close to the persistence adapter or query service. Do not leak SeaQuery types into core domain models unless the crate is explicitly a database adapter.
• Preserve parameter binding. Prefer APIs that build SQL plus values for the driver; reserve value-injected SQL strings for tests, sanitized logs, and debugging.
• Keep generated SQL reviewable. Name reusable predicates, avoid deeply nested builder chains, and snapshot only stable SQL shapes where that helps review.
• Make dialect choices explicit. PostgreSQL, SQLite, and MySQL/MariaDB differ in placeholders, quoting, upsert syntax, return clauses, JSON, arrays, date/time, locking, and DDL — and MySQL and MariaDB diverge from each other on some of these (for example, native JSON storage and RETURNING support) even under the same backend-mysql feature.

Prefer SeaQuery when:

• Query shape is genuinely dynamic.
• Many optional filters, sorts, joins, projections, or predicates must compose.
• The same query-building logic targets multiple SQL dialects.
• Repeated string concatenation would become unsafe or hard to review.
• The project already uses SeaQuery consistently.
• An AST-style builder improves maintainability.

Use SeaQuery with sqlx when SeaQuery builds the SQL and bind values while SQLx remains responsible for execution, pooling, transactions, migrations, and row mapping. In projects using sea-query-sqlx, inspect the installed version; current docs expose a SqlxBinder integration that builds SQLx-compatible SQL and values for sqlx::query_with.

Common commands:

sqlx migrate info
sqlx migrate run
sqlx migrate revert
cargo sqlx prepare --workspace -- --workspace --all-targets --features <supported-feature-set>
cargo sqlx prepare --check --workspace -- --workspace --all-targets --features <supported-feature-set>
SQLX_OFFLINE=true cargo check --workspace --all-targets --features <supported-feature-set>

Adapt command names to the repository's SQLx CLI version and recipes. Some projects use cargo sqlx ...; others install a standalone sqlx binary.

Schema And Migration Review

• Database invariants live in the database skill; the Rust review checks that Rust types, query mappings, migrations, and domain errors line up with them.
• SQLx migrations are invoked through the repository's chosen workflow. Do not assume sqlx migrate add/run/revert is authoritative when the repo has a custom migration generator or schema source of truth.
• Rust migration code, embedded migrations, and startup migration hooks have a rollback and deployment story. Avoid surprise DDL on application boot unless that is explicit repository policy.
• Seed data and fixtures are deterministic, isolated, and safe for repeated test runs.

SQL Library Choice

• Prefer raw SQL with SQLx macros when queries are static, readability matters, compile-time checking is available, and SQL expresses the behavior directly.
• Prefer SeaQuery when query shape is genuinely dynamic, many optional filters or dialects must compose, or string assembly would be unsafe or hard to review.
• Consider SeaORM only when the project already uses it or needs its entity, relation, migration, and active-model conventions across a broad persistence layer. Do not add it for one query or to avoid understanding SQL.
• Consider Diesel when the repository already uses Diesel or needs its typed query DSL, compile-time schema integration, and migration workflow. Do not mix it into a SQLx-centered project without a clear boundary.
• Prefer explicit raw SQL for migrations, performance-sensitive queries, database-native features, hand-tuned plans, or cases where an ORM/query builder hides important semantics.

SQLite Guidance

• Do not assume SQLite is a drop-in PostgreSQL substitute. Use sqlite-sql-engineering for SQLite-specific schema, transaction, PRAGMA, locking, and test-database behavior.
• Type affinity, foreign key enforcement, locking, datetime handling, DDL support, and concurrency behavior differ from PostgreSQL.
• Ensure foreign keys are enabled when tests or application logic rely on them.
• Keep migrations compatible with SQLite's DDL limits, or isolate SQLite-specific migrations when the repository already supports multiple database backends.
• Use SQLite for local or embedded workflows when that is the product choice, not as proof that PostgreSQL-specific SQL is correct.

MySQL/MariaDB Guidance

• Do not assume MySQL and MariaDB are interchangeable even though SQLx and SeaQuery share a single mysql/backend-mysql feature for both. Use mysql-mariadb-sql-engineering for engine-specific schema, storage-engine, replication, and query-plan behavior, and confirm which engine and version the code actually targets.
• SQLx's MySqlConnection connects to both engines, but engine-specific SQL (native JSON type behavior, window functions/CTEs on older MySQL, MariaDB RETURNING, INSERT ... ON DUPLICATE KEY UPDATE semantics) is not portable between them without verification.
• InnoDB is the default and correct storage engine for transactional tables on both; do not assume a schema imported from or generated against a non-InnoDB table (MyISAM, Aria, MEMORY) preserves transactional or foreign-key guarantees.
• Test against both engines in CI when the repository claims to support both; a passing MySQL integration test does not prove MariaDB compatibility, or the reverse.

SQL Review Checklist

• Queries select explicit columns and return only the data the caller needs.
• Predicates are sargable where possible and match available indexes.
• Pagination is stable under inserts/deletes; keyset pagination is considered for large ordered result sets.
• Joins preserve intended cardinality and do not accidentally multiply rows.
• NULL handling, collation, case sensitivity, timezone, and numeric precision are explicit where they affect behavior.
• Transactions cover exactly the consistency boundary and no more.
• User-controlled values are bound parameters, never interpolated SQL.
• Observability exists for slow queries, migration failures, and pool exhaustion when the repository has a logging/tracing pattern.
• For PostgreSQL-specific index, constraint, RLS, privilege, and query-plan review, load postgresql-sql-engineering.
• For MySQL/MariaDB-specific storage-engine, index, replication, and query-plan review, load mysql-mariadb-sql-engineering.
• For SQLite-specific PRAGMA, type-affinity, locking, migration, and temp database behavior, load sqlite-sql-engineering.

Testing And Acceptance Criteria

• Write a failing test or migration check before fixing a query bug when practical.
• Use BDD-style examples for user-visible persistence behavior, such as "Given an existing account, when a duplicate email is saved, then the caller receives a conflict and no partial write remains."
• Use integration tests against the database engine whose behavior matters.
• Verify migrations from an empty database and, when compatibility matters, from a representative previous schema.
• Run SQLx prepare/check commands after changing compile-checked queries, schema, or migrations.
• For SeaQuery builders, add focused tests for generated SQL shape and bind values when logic is dynamic. Cover no filters, one filter, multiple filters, sort order, pagination, empty inputs, NULL, and dialect-specific behavior.

Anti-Patterns

• Treating application validation as a substitute for database constraints on critical invariants.
• Returning raw SQLx or driver errors through domain or HTTP APIs.
• Building SQL by string concatenation with user input.
• Introducing SeaQuery for simple static SQL that would be clearer as sqlx macros or explicit SQL.
• Calling to_string or debug-rendered SQL with user-controlled values as an execution path instead of using bound values.
• Adding indexes speculatively without a query and write-cost rationale.
• Hiding long transactions behind generic repository helpers.
• Claiming PostgreSQL behavior from SQLite-only tests, or the reverse.
• Claiming MariaDB behavior from MySQL-only tests, or the reverse.