An agent that grades yesterday's agents

Your product runs hundreds of agent sessions a day: each one writing transcripts and outcomes to a sqlite URL. Quality drift is the silent killer: the agents start subtly worse, and nobody notices for a week.

A hoody-cron entry fires at 6am. It curls an evaluator agent with one prompt: "Read yesterday's sessions from the sqlite URL. Sample 50, score them on these rubrics, write findings to the report table, flag the worst three for human review."

The evaluator container wakes, does the work in 20 minutes, exits. By the time you're at your desk, you have a graded sample and three transcripts queued. The grader doesn't watch the agents. It runs on a cadence and judges them, like a teacher reading homework overnight.

The cron job is the supervisor; the supervisor is also an agent.