Your Coding Agent Doesn't Need Better Prompts. It Needs a Contract.

How I structured a repo to make AI agent drift visible before it ships. The most dangerous failure mode I've seen in agentic coding workflows is not broken code. Broken code is at least visible. The dangerous failure is plausible code: code that passes tests, implements something close to the request, and quietly expands the product surface in a direction nobody approved. No bug. No crash. Just drift. After a few months of fighting this, I stopped trying to write better prompts. The issue wasn't