I built LOLM, a lower-cost LLM agent that shows what it actually did — looking for blunt feedback

Structured task state under the token stream — not just chat. Control that blocks false “done.” Receipts you can audit.