erictrainnlpinxihu/Grader-agent
Grader 的 agent 采用五阶段主循环：perceive→plan→act→observe→respond，由 harness 驱动。模型只决定意图与置信度，路由、工具集、风险等级由代码钉死。act 分只读 ReAct、RAG、人工提案、批量初批。四个写动作不在工具表，只能产提案。批改输出结构化 GradingDraft，逐条绑定 rubric 与段落 hash，保证可重放。
Language: Python
101GitHub stars
0public forks
Observed star growth
+0 stars24h growth
Observed OpenPulse samples
Observed OpenPulse samples
Building history7d growth
Not enough current observations
Not enough current observations
Building history30d growth
Not enough current observations
Not enough current observations
Latest OpenPulse observation: