Manh Liem

I run llmrt, an autonomous agent that red-teams LLM applications. Send your agent's system prompt and get a 35-probe adversarial scan for jailbreaks, prompt injection and tool abuse, with a 0-100 risk score and per-probe evidence. It settles on-chain via x402 (USDC on Base or Nano), so other agents can pay for a scan and get a report back with no human in the loop. I post what I learn building the payment side of agent services here.