signalHacker News Show HN2026-09-30
Show HN: Rules that stop AI agents from wrecking your warehouse, with an eval
Claude Haiku 4.5 with no rules deleted every row of prod.orders, divided all 500 prices by 10 when only 60 were wrong, and invented countries for 52 customers. The proposed eight rules, five skills, a PII-masking preview tool and a query cost check prevented all such failures across 27 runs. An eval runs locally against DuckDB to verify the claims.
- for who
- Teams using AI coding agents to write SQL or manage data warehouses
- why now
- New rules and eval prevent AI agents from deleting production data, timely with Claude Haiku 4.5.
- what changes
- AI agents no longer scan, overwrite, or invent data without checks, making warehouse operations safer
- to do
- Install the data-agent-rules plugin or append the eight rules to your AGENTS.md file
key points
- Eight rules cover SQL, Spark, dbt, Databricks, Snowflake, BigQuery and DuckDB
- Rules include look-before-scan, no invented values, dev-only writes, and prove-changes
- safe_peek.py masks personal data and a cost guard limits BigQuery scans
#ai agent safety#data protection rules#sql protection#eval tool
score
score 8 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source