Sql Lint

SQL Lint

Use this when a project has a SQL catalog (-- name: X :type comments or any named-query convention) and wants to enforce basic hygiene without adopting a heavy SQL linter.

Two layers

This skill ships two linters, with different cost / coverage trade-offs.

Layer 1: catalog-lint (built-in, zero deps)

scripts/catalog-lint.mjs is a Node 20+ script with no dependencies. It catches:

• duplicate-name: two queries with the same -- name: X header. Some codegens silently shadow; others break.
• missing-semicolon: queries that omit the trailing ;. sqlc rejects these at build time, but catalog-lint surfaces them in fmt-time.
• empty-body: header with no SQL body.
• malformed-name: -- name: line that doesn't match name: X :type.
• select-star: SELECT * (always over-fetch).
• double-wildcard-like: LIKE '%...%' (full-table scan risk).

Run (this is the default first pass — start here before reaching for sqlfluff):

bashCopy code
# `scripts/` is relative to THIS skill's directory. From elsewhere, use the
# absolute path, e.g. node "$CLAUDE_SKILLS_DIR/sql-lint/scripts/catalog-lint.mjs".
node scripts/catalog-lint.mjs your-project/db/queries.sql
echo "exit=$?"   # 0 = clean, 1 = findings — the exit code IS the pass/fail signal

Exits 0 on clean / 1 on findings. Output is file:line: [rule] message, parseable by editor diagnostic gutters. Note: for select-star and double-wildcard-like, the reported line is the query's -- name: header, not the offending SQL line.

Layer 2: sqlfluff (optional, opt-in)

For full style coverage — keyword case, indentation, alias names, etc. — use sqlfluff. The skill ships .sqlfluff tuned for SQLite/D1 + sqlc catalogs:

bashCopy code
uvx sqlfluff lint --config .sqlfluff your-project/db/queries.sql

uvx (from astral.sh/uv) runs sqlfluff in an ephemeral venv; nothing is installed permanently. If uv is missing, fall back to pip install sqlfluff or pipx install sqlfluff.

The shipped config disables a few sqlfluff rules:

• AM04 (SELECT *) is handled by catalog-lint with better context.
• CP02 (column case) is too noisy on JSON path references like obj.field_name.
• L009 (trailing newline) is enforced by git.

When to invoke

• Every commit that touches the query catalog (pre-commit / pre-push hook).
• Once per migration milestone, run sqlfluff for the full style pass.

Inline SQL in host code

This skill does not police inline SQL in MoonBit / Rust / Go / etc. — ast-grep is the right tool there, but it requires per-language tree-sitter support. For host languages without a tree-sitter grammar (e.g. MoonBit), a text-grep baseline file (.linters/inline-sql-baseline.json) works: count .prepare( occurrences per file and refuse increases.

The companion sql-security skill documents the same approach for SQL injection.

CI integration

justCopy code
sql-lint:
    node scripts/catalog-lint.mjs db/queries.sql

For full sqlfluff in CI, run a separate sql-lint-style step so the cheap catalog-lint can fail fast.

Engine extensibility

catalog-lint.mjs is engine-agnostic — the rules look at SQL comments, semicolons, and basic patterns. Adapt to MySQL / Postgres by changing the sqlfluff dialect = sqlite line in .sqlfluff.

Requirements

• Node 20 or newer for scripts/catalog-lint.mjs.
• (Optional) uv / pipx / pip for sqlfluff.

Files

• scripts/catalog-lint.mjs — zero-dep catalog linter.
• .sqlfluff — opt-in sqlfluff config (SQLite dialect; change dialect = ... for other engines).