DB Read-Only SQL

Use this skill for repository-scoped database lookups.

Rules

• •Use read-only SQL only (SELECT or WITH ... SELECT).
• •Never run mutation or schema-changing SQL.
• •Never run multi-statement SQL from user input.
• •Confirm table and column names from packages/db/prisma/schema.prisma before drafting SQL.
• •Draft the SQL and exact psql command first.
• •Ask the user to confirm before executing.

Schema Discovery

• •First source of truth: packages/db/prisma/schema.prisma.
• •If table/column names are still uncertain, draft and run an information_schema lookup query first, then draft the final query.

Disallowed Keywords

Reject SQL containing these case-insensitive keywords:

• •insert
• •update
• •delete
• •drop
• •alter
• •create
• •truncate
• •grant
• •revoke
• •copy
• •call
• •do
• •vacuum
• •analyze

Also reject semicolon-chained statements.

Default Query Limits

• •If query is not an aggregate-only result and has no LIMIT, append LIMIT 50.
• •If LIMIT exceeds 500, reduce to 500.

Required Draft Step

Before any execution, provide:

1. •Draft SQL
2. •Exact command to run
3. •Short expected output note

Then ask: Run this query? (yes/no)

Do not execute until the user confirms.

Execution Command

Preferred command (loads repo .env first, then runs read-only transaction):

set -a; source .env >/dev/null 2>&1; set +a; psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -X -P pager=off -c "BEGIN READ ONLY; <SQL>; ROLLBACK;"

For single-value counts, prefer machine-friendly output:

set -a; source .env >/dev/null 2>&1; set +a; psql "$DATABASE_URL" -v ON_ERROR_STOP=1 -X -At -c "SELECT COUNT(*)::int FROM <table>;"

Environment Checks

Before execution:

• •Confirm DATABASE_URL is set after loading .env.
• •If still missing, check fallback env files:
  • •apps/web/.env
  • •apps/chainworker/.env

If connection fails, run a quick host diagnostic:

node -e "const u=new URL(process.env.DATABASE_URL||''); console.log(u.host)"

If DNS/network fails in sandbox, rerun the query outside sandbox (with approval) rather than changing SQL.

Output Style

• •Return a concise summary first.
• •Include key rows/values from the result.
• •Include the SQL that was run in a code block.