数学 agent llm github workflow developer-tools paper memory llms arxiv 工具 mcp rag coding-agent tool-use multi-agent daily-paper 开发 推荐系统 机器学习 语音识别 daily-pick nlp no-pick reinforcement-learning sandbox skills 搜索 深度学习 benchmark observability claude-code codex orchestration reliability retrieval runtime spark verification 开发语言 AI Agent Multi-Agent agents ai-coding bert browser computer-use deer-flow eval gui-agent infra llm-systems multimodal planning reasoning safety 信息抽取 数据结构 Benchmark LLM LLM Agent Tool Use archon automation cli code-intelligence coordination data deep-research devtools evaluation hallucination hermes-agent hierarchy inference knowledge-graph langchain orgagent project python research-agent speculative-decoding token-efficiency