REAP turns real coding-agent use into automatic benchmarks

REAP is presented as a way to automatically build for from real interactive . The main idea is to turn the work a already sees in live use into evaluation material, instead of relying only on hand-made test tasks. The available information does not confirm numbers, cost savings, setup steps, or whether the code is .

Key points

  • REAP is about automatically creating .
  • The source material comes from interactive .
  • The goal appears to be reducing manual work in making evaluation tasks.
  • It is more relevant to agent evaluation than to direct token savings.
Read original