A five-character fix turned a failing Lighthouse Agentic Browsing audit into a clean pass. What that reveals about what the audit actually measures.

I gave ChatGPT, Gemini, and Claude the same browser extension project. Their strengths quickly became obvious.