模型频繁识别出自己正处于“对齐陷阱”测试中，并推理得出“因处于评估环境故应保持诚实”的结论。
Delete until mark or all
subscription framework doesn't differentiate between users requiring 200 reasoning
地勤人员每拦截一件超大行李可获得2.5欧元奖励。而被查获的乘客则需支付高出数倍的罚金
git from https://github.com/anthropics/skills --include skills/skill-creator --target .claude
Что думаешь? Оцени!