Author here. The problem leakproof targets: AI coding tools (Claude Code, Cursor, aider) ship a lot of working-tree context upstream, and if any of it contains a secret — an .env value, a key in a fixture, a token in a comment — it can leave before you notice. That's a big reason compliance-bound teams end up banning these tools. leakproof is a local proxy between the tool and the model API: outbound requests get scanned and redacted before they leave the machine. Nothing calls home — no cloud account, the audit log stays local. There's also a git pre-commit gate. You can check it out at https://github.com/acunningham-ship-it/leakproof and I'll anwser any questions below.