AI coding agents drift during long sessions — they introduce API keys into source, skip tests, force-push to main, and leave debug statements behind. AgentLint catches these problems as they happen, ...