30-second scan
Are you a relay?
Six statements. Answer honestly — nobody sees this but you. At the end you get a rough position on the L1–L7 ladder.
0 of 6—
I paste the agent's answer into a PR without opening the diff.
When someone asks me something, I forward the question to the model in the same words.
My answer to a colleague starts with "the AI says…".
If the model were down today, I could not explain why my own code works.
My whole toolchain is one chat tab.
I find out about a bug from a reviewer, not from my own tests.