Opus 4.6, high effort, detailed spec. Complex order fulfillment system — state machines, shipment groups, payments, lost package handling. 30 min later: working app. Hundreds of thousands of property tests. All green. Then we ran formal proofs. Three bugs.

Hi bsky! We build tools that prove your AI code is correct. We use formal verification and publish open source tools. Our first tool is lemmafit. An npm package you can use with Claude Code to run proofs on logic. You don't have to know formal verification at all.