I got paranoid about OpenClaw skills injecting crap into my system prompt, so I built a quarantine pipeline with two LLMs as reviewers (93.75% detection, zero false negatives)
8/10The author built a quarantine system using two LLMs as reviewers to analyze OpenClaw skills, achieving 93.75% detection accuracy with zero false negatives. This system aims to prevent malicious code injections into the OpenClaw system prompt, which currently lacks sanitization.
