Skill: Review SQL

Purpose

Use Cases

• Orchestrated review: Used as the language step when review-code runs for projects that include SQL (.sql files, embedded SQL, or ORM-generated SQL).
• SQL-only review: When the user wants only query correctness, performance, and safety checked.
• Migration or portability: Check dialect-specific constructs and portability across databases.

Behavior

Scope of this skill

• Analyze: SQL and query logic in the given scope (files, snippets, or diff). Accept .sql files, embedded SQL in application code, or ORM-generated SQL when visible.
• Do not: Decide scope (diff vs codebase); do not perform full application security or architecture review. Focus on SQL/query dimension.

Review checklist (SQL dimension only)

1. Injection and parameterization: No string concatenation or interpolation for user input in SQL; use parameterized queries or prepared statements; avoid dynamic SQL from untrusted input.
2. Indexing and execution plan: Queries that filter or join on unindexed columns; SELECT * on large tables; missing indexes for WHERE/JOIN/ORDER BY.
3. Transactions and isolation: Appropriate transaction boundaries; isolation level and locking; avoid long-running transactions; deadlock risk.
4. NULL and unique constraints: Handling of NULL in comparisons and aggregates; unique constraints and duplicate handling; NOT NULL where appropriate.
5. Dialect and portability: Database-specific syntax (e.g. LIMIT vs OFFSET/FETCH, date functions) and portability if multi-DB support is required.
6. Large tables and paging: Full scans on large tables; paging (keyset vs OFFSET) and scalability.
7. Sensitive columns and permissions: Sensitive data in SELECT; least-privilege and role usage in SQL (where visible).

Tone and references

• Professional and technical: Reference specific locations (file:line or query identifier). Emit findings with Location, Category, Severity, Title, Description, Suggestion.

Input & Output

Input

• Code scope: Files or snippets containing SQL (e.g. .sql files, code with embedded SQL, or ORM-generated SQL when available). Provided by the user or scope skill.

Output

• Emit zero or more findings in the format defined in Appendix: Output contract.
• Category for this skill is language-sql.

Restrictions

• Do not perform scope selection or full security/architecture review. Stay within SQL and query conventions.
• Do not give conclusions without specific locations or actionable suggestions.
• Do not assume a specific database vendor unless stated; note dialect when relevant.

Self-Check

• Was only the SQL/query dimension reviewed (no scope/architecture beyond query design)?
• Are parameterization, indexing, transactions, NULL/constraints, and portability covered where relevant?
• Is each finding emitted with Location, Category=language-sql, Severity, Title, Description, and optional Suggestion?
• Are issues referenced with file:line or query identifier?

Examples

Example 1: String concatenation in query

• Input: Query built with string concatenation including user input.
• Expected: Emit a critical finding for SQL injection; suggest parameterized query or prepared statement. Category = language-sql.

Example 2: Large table without paging

• Input: SELECT * FROM large_table ORDER BY id without LIMIT or paging.
• Expected: Emit finding for performance and scalability; suggest paging (e.g. keyset or OFFSET/FETCH) and avoid SELECT * if not needed. Category = language-sql.

Edge case: ORM-generated SQL

• Input: Only application code using an ORM; generated SQL not visible.
• Expected: Review any raw SQL or query builders in the code; if no SQL is visible, state that and skip or report "No SQL to review in scope." Do not invent SQL.

Appendix: Output contract

path/to/file.ext

language-sql

critical

major

minor

suggestion

- **Location**: `scripts/orders.sql:12`
- **Category**: language-sql
- **Severity**: critical
- **Title**: Query built with string concatenation; injection risk
- **Description**: User-controlled input is concatenated into the WHERE clause.
- **Suggestion**: Use parameterized query or prepared statement with placeholders.