The verified text-to-SQL blog
Ask Postgres in plain English.Trust the SQL.
Guides for teams who want a text-to-SQL AI agent they can actually trust: every query read-only, every answer traceable to the SQL that produced it, and the whole skill portable to Claude, Cursor, and Codex.
Read-only SELECT. Row caps enforced. Your RLS policies honored. Every chart cites its query. No agent to build.
See how Chion worksRun a text-to-SQL agent your data team trusts
agent skills
Claude Code Data Analyst: One Skill per Role
Turn Claude Code into a data analyst workforce: one skill per role, answering from your database with a chart and the verified, read-only SQL behind it.
ai sql
SKILL.md SQL Examples: Claude Code, Codex, Cursor
See a real SQL SKILL.md example (YAML frontmatter plus one verified SELECT) and run the same skill file in Claude Code, Codex, Cursor, and Chion.
ai agent configuration
CLAUDE.md for SQL: Make Claude Run Verified Queries
CLAUDE.md for SQL: keep Claude Code on your conventions and pair the file with a portable SKILL.md bundle of verified, read-only Postgres queries.
agent skills
Claude Code Skill for SQL: Survives Schema Changes
Claude Code skills for Postgres break when schemas change. Get the SKILL.md YAML frontmatter right and ground the body in verified, read-only queries.
ai sql
SQL Agent Skills: Reuse Your Trusted Queries
A SQL agent skill is a verified read-only SELECT your team trusts, compiled once into a reusable skill any AI agent runs the same way every time.
ai sql
AI SQL Agent in Production: 4 Required Properties
The four properties an AI SQL agent must prove before production: read-only enforced in code, a credential vault, an audit trail, and verified SQL skills.
ai analytics
What is an AI data analyst? Verified SQL on Postgres
What an AI data analyst does: turn plain-English questions into a chart and the verified, read-only SQL it ran on your own Postgres, and how to trust one.
data architecture
Semantic Layer for SQL: How AI Gets Verified Answers
A semantic layer for SQL maps each business term to its exact table, column, and filter, so the same prompt resolves to the same verified, read-only SQL.
ai sql
Text-to-SQL on Postgres: Hallucinations and the Fix
How text-to-SQL works on Postgres: the five pipeline stages, why hallucinated joins and grain errors happen, and how read-only verified SQL skills fix re-asks.
claude-code
Claude Code says "SKILL.md must start with YAML frontmatter (---)": the exact fix
Fix the Claude Code error SKILL.md must start with YAML frontmatter (---): the valid shape plus five ways files fail it, with broken vs fixed examples.
security
Stopping prompt injection in a text-to-SQL agent: the layers that actually catch it
How Chion stops text-to-sql prompt injection (OWASP LLM01): a 400-char input sanitizer, a read-only AST validator, and a 1,000-row capped database role.