Proof Before Reuse: showing agents only what later sessions reproduced keeps a poisoned skill out of procedural memory [R]

Explain Like I'm Normal

Researchers developed a 'Proof Before Reuse' system to prevent autonomous agents from learning malicious code patterns from single 'poisoned' sessions. The system only commits a new skill to long-term memory if multiple subsequent sessions independently reproduce the same logic and pass verification. This creates a filter that strips away one-off malicious injections while retaining legitimate procedural improvements.