Nature, Published online: 11 March 2026; doi:10.1038/s41586-026-10216-0
A recent paper from ETH Zürich evaluated whether these repository-level context files actually help coding agents complete tasks. The finding was counterintuitive: across multiple agents and models, context files tended to reduce task success rates while increasing inference cost by over 20%. Agents given context files explored more broadly, ran more tests, traversed more files — but all that thoroughness delayed them from actually reaching the code that needed fixing. The files acted like a checklist that agents took too seriously.
，详情可参考WhatsApp Web 網頁版登入
В рыболовной сети нашли 15-метровую тушу редкого кита20:45。谷歌是该领域的重要参考
В музее стали предлагать понюхать навоз редких коней ради их спасенияВ Японии в музее стали предлагать понюхать конский навоз ради исчезающих лошадей