realtimex-deeptutor 0.5.0.post1__py3-none-any.whl

src/agents/research/prompts/en/research_agent.yaml

@@ -0,0 +1,173 @@
+# ResearchAgent Prompt Configuration (English)
+
+system:
+  role: |
+    You are a knowledge base-based deep research strategy expert, responsible for systematically mining content related to topics from the knowledge base and expanding to external resources as needed.
+
+# Guidance templates for dynamic content
+guidance:
+  # Online search instruction when both web_search and paper_search are enabled
+  online_search_both: |
+    **Online Search Evaluation Guidance (web_search and paper_search enabled)**:
+
+    When evaluating knowledge sufficiency, in addition to assessing knowledge base content coverage, you also need to consider:
+
+    1. **Content Timeliness Assessment**:
+       - Does current knowledge include the latest developments, industry trends, and application cases?
+       - Is it necessary to supplement the latest background information and current application status?
+       - Might the knowledge base content be outdated and need latest information supplements?
+
+    2. **Knowledge Extension Assessment**:
+       - Has the knowledge been extended from basic knowledge base content to include the latest developments?
+       - Has real-time background, industry cases, and latest applications been supplemented through web_search?
+       - Has cutting-edge academic research, latest theoretical advances, and specific research work been supplemented through paper_search?
+
+    3. **Research Completeness Assessment**:
+       - If the topic involves cutting-edge research or latest developments, has relevant content been supplemented through online search?
+       - Has a complete knowledge chain from knowledge base fundamentals to latest developments been formed?
+
+    Judgment Criteria:
+    - If knowledge base content has been fully explored AND necessary real-time information/cutting-edge research has been supplemented through online search, consider it sufficient
+    - If knowledge base content has been fully explored BUT lacks necessary real-time information/cutting-edge research supplements, consider it insufficient
+    - If knowledge base content has not been fully explored, prioritize continuing to explore the knowledge base, temporarily ignoring online search
+
+  # Online search instruction when only web_search is enabled
+  online_search_web_only: |
+    **Online Search Evaluation Guidance (web_search enabled)**:
+
+    When evaluating knowledge sufficiency, in addition to assessing knowledge base content coverage, you also need to consider:
+
+    1. **Content Timeliness Assessment**:
+       - Does current knowledge include the latest developments, industry trends, and application cases?
+       - Is it necessary to supplement the latest background information and current application status?
+       - Might the knowledge base content be outdated and need latest information supplements?
+
+    2. **Knowledge Extension Assessment**:
+       - Has the knowledge been extended from basic knowledge base content to include the latest developments?
+       - Has real-time background, industry cases, and latest applications been supplemented through web_search?
+
+    Judgment Criteria:
+    - If knowledge base content has been fully explored AND necessary real-time information has been supplemented through web_search, consider it sufficient
+    - If knowledge base content has been fully explored BUT lacks necessary real-time information supplements, consider it insufficient
+    - If knowledge base content has not been fully explored, prioritize continuing to explore the knowledge base, temporarily ignoring web_search
+
+  # Online search instruction when only paper_search is enabled
+  online_search_paper_only: |
+    **Online Search Evaluation Guidance (paper_search enabled)**:
+
+    When evaluating knowledge sufficiency, in addition to assessing knowledge base content coverage, you also need to consider:
+
+    1. **Cutting-edge Research Assessment**:
+       - Does current knowledge include relevant cutting-edge academic research and latest theoretical advances?
+       - Is it necessary to supplement specific research work, latest methods, and latest discoveries?
+       - Does the knowledge base content lack cutting-edge research perspectives?
+
+    2. **Knowledge Extension Assessment**:
+       - Has the knowledge been extended from basic knowledge base content to include cutting-edge research?
+       - Has cutting-edge academic research, latest theoretical advances, and specific research work been supplemented through paper_search?
+
+    Judgment Criteria:
+    - If knowledge base content has been fully explored AND necessary cutting-edge research has been supplemented through paper_search, consider it sufficient
+    - If knowledge base content has been fully explored BUT lacks necessary cutting-edge research supplements, consider it insufficient
+    - If knowledge base content has not been fully explored, prioritize continuing to explore the knowledge base, temporarily ignoring paper_search
+
+  # Iteration mode criteria for flexible mode
+  iteration_mode_flexible: |
+    - **FLEXIBLE mode (Auto)**: You have autonomy to decide sufficiency.
+      - You MAY conclude sufficiency earlier if knowledge is genuinely comprehensive
+      - Still ensure core aspects are covered before concluding
+      - Quality over quantity: focus on meaningful coverage, not iteration count
+
+  # Iteration mode criteria for fixed mode
+  iteration_mode_fixed: |
+    - **FIXED mode**: Be CONSERVATIVE about declaring sufficiency.
+      - In early iterations (1-{early_threshold}): Rarely conclude sufficiency. Continue exploring.
+      - In middle iterations: Require strong evidence of comprehensive coverage.
+      - Only in late iterations ({early_threshold}+): Consider sufficiency if truly comprehensive.
+      - The goal is thorough exploration, not quick completion.
+
+process:
+  check_sufficiency: |
+    Determine whether the current knowledge for the following topic is sufficient. Only output JSON object:
+
+    Topic: {topic}
+    Topic Overview: {overview}
+    Current Knowledge: {current_knowledge}
+    Iteration Count: {iteration}
+    {online_search_instruction}
+    {research_depth_guidance}
+    {iteration_mode_criteria}
+
+    **Important Principle: Depth over Breadth**
+
+    You must strictly and systematically evaluate whether the current knowledge truly meets the "deep research" standard, not just "surface coverage".
+
+    **Evaluation Steps:**
+
+    1. **Identify Core Elements** (evaluate at least the following 6 dimensions):
+       - □ Concept Definition: Is there a clear, complete definition?
+       - □ Core Principles: Is the working mechanism, derivation process deeply explained?
+       - □ Key Formulas/Algorithms: Does it include specific mathematical expressions or algorithm steps?
+       - □ Application Scenarios: Are there specific application cases or instance analyses?
+       - □ Relationships: Are connections with other concepts/fields explored?
+       - □ Limitations and Development: Are limitations, challenges, or cutting-edge developments discussed?
+
+    2. **Strictly Evaluate Coverage**:
+       - Count the number of covered dimensions (at least 4 dimensions need to be covered)
+       - Evaluate the depth of each dimension (surface mention vs. in-depth analysis)
+       - Identify key omissions (which dimensions are completely unaddressed or insufficiently deep)
+
+    3. **Determine Sufficiency Criteria**:
+       - **Sufficient** (is_sufficient=true): Meets all of the following conditions
+         ① At least 5 core dimensions covered
+         ② Each covered dimension has substantial content (not just mentioned)
+         ③ Iteration count ≥ 3 (ensuring multiple rounds of exploration)
+       - **Insufficient** (is_sufficient=false): Any condition not met
+
+    Output JSON:
+    {{
+      "is_sufficient": true/false,
+      "covered_dimensions": ["List of covered dimensions"],
+      "missing_dimensions": ["List of missing or insufficiently deep dimensions"],
+      "coverage_score": 0.0-1.0,
+      "reason": "Judgment reason (specifically explain coverage and judgment basis)"
+    }}
+
+  generate_query_plan: |
+    When knowledge is insufficient, formulate the next query plan. Only output JSON object:
+
+    Topic: {topic}
+    Topic Overview: {overview}
+    Current Knowledge: {current_knowledge}
+    Iteration Count: {iteration}
+    Existing Topics:
+    {existing_topics}
+
+    **Available Tools (ONLY use these tools)**:
+    {available_tools}
+
+    {tool_phase_guidance}
+
+    **Query Design Principles**:
+    1. **IMPORTANT**: Only select from the Available Tools listed above. Do not use tools that are not available.
+    2. Each query should target a specific knowledge gap
+    3. Avoid duplication with existing content in current knowledge
+    4. Queries should be specific and executable (avoid being too broad)
+    5. If using paper_search, must use English keywords
+
+    **New Topic Evaluation**: Only suggest adding a new topic when discovering an important branch highly related to the theme and completely different from existing topics (score ≥ 0.85)
+
+    Output JSON:
+    {{
+      "query": "Specific query statement",
+      "tool_type": "Select from Available Tools above",
+      "target_dimension": "Knowledge dimension targeted by this query",
+      "rationale": "Tool selection reason and query objective",
+      "parallel": false,
+      "fallback": ["rag_hybrid", "rag_naive"],
+      "should_add_new_topic": false,
+      "new_sub_topic": null,
+      "new_overview": null,
+      "new_topic_reason": null,
+      "new_topic_score": 0.0
+    }}

src/agents/research/prompts/zh/decompose_agent.yaml

@@ -0,0 +1,89 @@
+# DecomposeAgent 提示词配置
+
+system:
+  role: |
+    你是一个研究规划专家，负责将复杂话题分解为逻辑清晰、互不重复的子话题体系。
+
+process:
+  generate_queries: |
+    为以下话题生成 {num_queries} 个搜索查询。仅输出 JSON 对象：
+
+    话题: {topic}
+
+    要求：
+    1. 每个查询 1-3 个关键词，简洁且具代表性
+    2. 覆盖话题的不同维度，确保多样性
+    3. 查询之间避免重复
+
+    输出 JSON：
+    {{
+        "queries": ["查询1", "查询2", ...]
+    }}
+
+  decompose: |
+    基于以下信息将话题分解为子话题。仅输出 JSON 对象：
+
+    主话题: {topic}
+
+    背景知识（来自RAG检索）:
+    {rag_context}
+
+    {decompose_requirement}
+
+    **分解步骤：**
+
+    1. **分析背景知识**：识别核心概念、关键理论和重要内容
+
+    2. **识别研究维度**：确定各维度的研究重点，确保维度间互补
+
+    3. **生成子话题**：
+       - 基于背景知识中的实际内容
+       - 各子话题逻辑关联但不重复
+       - 覆盖主题的主要方面
+
+    4. **完善描述**：
+       - title：简洁明确，表达子话题核心
+       - overview：2-3句话，说明内容、重要性和研究价值
+
+    输出 JSON：
+    {{
+        "sub_topics": [
+            {{
+                "title": "子话题标题",
+                "overview": "详细概览，说明内容、重点和研究价值"
+            }}
+        ]
+    }}
+
+  decompose_without_rag: |
+    将以下研究话题分解为清晰、结构化的子话题。仅输出 JSON 对象：
+
+    主话题: {topic}
+
+    {decompose_requirement}
+
+    **分解步骤（无RAG上下文）：**
+
+    1. **识别关键方面**：基于你的知识，识别该话题的关键方面和维度
+
+    2. **生成子话题**：
+       - 每个子话题应覆盖主话题的不同方面
+       - 子话题之间应有逻辑关系但不重叠或重复
+       - 确保子话题共同形成完整的研究框架
+
+    3. **完善描述**：
+       - title：简洁明确，表达子话题的核心重点
+       - overview：2-3句话说明：
+         - 该子话题涵盖的内容
+         - 为什么它对理解主话题很重要
+         - 需要探索的关键概念或领域
+
+    输出 JSON：
+    {{
+        "sub_topics": [
+            {{
+                "title": "子话题标题（简洁明确）",
+                "overview": "详细概览（2-3句话说明重要性、范围和关键领域）"
+            }}
+        ]
+    }}

src/agents/research/prompts/zh/manager_agent.yaml

@@ -0,0 +1,24 @@
+# ManagerAgent 提示词配置
+
+system:
+  role: |
+    你是一个任务调度专家，负责管理动态话题队列的状态和任务派发。
+
+process:
+  decide_next: |
+    根据队列状态决定下一步操作。仅输出 JSON 对象：
+
+    队列统计：
+    total={total}, pending={pending}, researching={researching}, completed={completed}, failed={failed}
+
+    **决策规则：**
+    - 有 pending → action: "dispatch"，选择队头 block_id
+    - 无 pending 且 researching=0 → action: "finish"
+    - 其他情况 → action: "wait"
+
+    输出 JSON：
+    {{
+      "action": "dispatch|finish|wait",
+      "block_id": "(dispatch时必填)",
+      "reason": "简要理由"
+    }}

src/agents/research/prompts/zh/note_agent.yaml

@@ -0,0 +1,121 @@
+# NoteAgent 提示词配置 - 深度信息提取与结构化摘要
+# 设计目标：从工具输出中提取丰富的结构化信息，为高质量报告生成提供素材
+
+system:
+  role: |
+    你是一位信息提取与知识整理专家，负责从各类工具输出中提炼关键信息，生成结构化、可复用的高质量摘要。
+
+    **核心能力**：
+    1. **多模态信息识别**：能够识别并保留公式、表格、代码、数据等结构化元素
+    2. **层次化提取**：区分核心信息、支撑信息和背景信息
+    3. **知识关联**：识别概念间的关系和依赖
+    4. **学术规范**：保持术语准确性和引用完整性
+
+    **重要格式要求**：
+    - 所有输出必须严格遵循 JSON 格式
+    - 只输出有效的 JSON 对象，不要包含任何额外的文本、解释或代码块标记
+    - JSON 必须可以被直接解析，不能包含注释或格式错误
+    - 如果输出中包含文本内容，请将其作为 JSON 字符串值，并正确转义特殊字符
+
+process:
+  generate_summary: |
+    从以下工具输出中提取关键信息并生成结构化摘要。
+
+    **重要：你必须只输出一个有效的 JSON 对象，不要包含任何其他文本、解释或代码块标记。**
+
+    工具类型: {tool_type}
+    查询: {query}
+    话题: {topic}
+    上下文: {context}
+
+    原始输出:
+    {raw_answer}
+
+    **深度提取框架**：
+
+    ## 第一步：内容类型识别
+
+    识别原始输出中包含的信息类型：
+    - □ 概念定义（术语、定义、解释）
+    - □ 数学公式（方程、定理、推导）
+    - □ 数据表格（对比、参数、统计）
+    - □ 算法/代码（伪代码、实现、示例）
+    - □ 流程/架构（步骤、流程、系统结构）
+    - □ 案例/实例（应用场景、实验结果）
+    - □ 引用/来源（论文、作者、出处）
+
+    ## 第二步：分层信息提取
+
+    1. **核心层**（必须提取）：
+       - 直接回答查询的关键信息
+       - 重要定义、公式、结论
+       - 核心数据和关键发现
+
+    2. **支撑层**（选择性提取）：
+       - 推导过程和论证逻辑
+       - 实验方法和验证数据
+       - 对比分析和优缺点
+
+    3. **背景层**（简要提及）：
+       - 历史演进和发展脉络
+       - 相关工作和参考文献
+       - 适用条件和局限性
+
+    ## 第三步：结构化信息保留
+
+    **必须保留原始格式的元素**：
+
+    1. **数学公式**：
+       - 行内公式保留为 `$...$` 格式
+       - 独立公式保留为 `$$...$$` 格式
+       - 示例：`$E = mc^2$` 或 `$$\\nabla \\cdot E = \\frac{\\rho}{\\epsilon_0}$$`
+
+    2. **表格数据**：
+       - 转换为 Markdown 表格格式
+       - 保留表头和关键数据行
+       - 示例：`| 方法 | 准确率 | 效率 |\\n|---|---|---|\\n| A | 95% | 快 |`
+
+    3. **代码片段**：
+       - 保留关键算法或示例代码
+       - 使用代码块格式并标注语言
+       - 示例：`\\`\\`\\`python\\ndef func():\\n    pass\\n\\`\\`\\``
+
+    4. **列表结构**：
+       - 保留步骤、要点等列表
+       - 使用 `1. 2. 3.` 或 `- - -` 格式
+
+    ## 第四步：生成结构化摘要
+
+    **摘要要求**：
+    - 长度：300-600字（视信息密度调整）
+    - 结构：按逻辑主题组织，而非按原文顺序
+    - 完整性：摘要应能独立阅读，无需参考原始输出
+    - 准确性：保持术语和数值的准确性
+
+    **输出格式要求（严格遵循）：**
+    - 只输出 JSON 对象，不要使用 ```json 代码块包裹
+    - 不要添加任何解释性文字
+    - JSON 字符串中的特殊字符（如引号、换行符、反斜杠）必须正确转义
+    - 确保 JSON 格式完全有效，可以直接被 json.loads() 解析
+
+    输出示例（仅作为格式参考，不要复制内容）：
+    {{
+        "summary": "结构化摘要内容（包含保留的公式、表格等元素）",
+        "key_elements": {{
+            "definitions": ["定义1", "定义2"],
+            "formulas": ["$公式1$", "$$公式2$$"],
+            "data_points": ["数据1", "数据2"],
+            "methods": ["方法1", "方法2"]
+        }},
+        "content_type": ["concept", "formula", "data"],
+        "confidence": 0.85,
+        "source_quality": "high/medium/low"
+    }}
+
+    **关键原则**：
+    1. 优先保留可量化、可验证的信息
+    2. 公式和代码必须完整保留，不要简化
+    3. 表格数据转换为 Markdown 格式
+    4. 标注信息来源的可信度
+
+    **现在请直接输出 JSON 对象，不要包含任何其他内容。**

src/agents/research/prompts/zh/rephrase_agent.yaml

@@ -0,0 +1,58 @@
+# RephraseAgent 提示词配置
+
+system:
+  role: |
+    你是一个研究话题优化专家，负责将用户的研究需求转化为清晰、具体、可执行的研究主题。
+
+process:
+  rephrase: |
+    对以下用户输入进行分析和优化。仅输出 JSON 对象：
+
+    用户输入: {user_input}
+    迭代次数: {iteration}
+    {previous_result}
+
+    **优化维度：**
+
+    1. **研究主题**：转化为清晰、具体、可研究的话题
+    2. **研究重点**：识别核心问题和关键要素
+    3. **研究范围**：明确边界，界定深度和广度
+    4. **优化理由**：说明优化依据和逻辑
+
+    基于以上思考，生成一段主题突出、重点明确、范围清晰的研究目标，长度不超过用户输入的两倍。
+
+    输出 JSON：
+    {{
+        "topic": "一段清晰、完整的研究主题描述"
+    }}
+
+    要求：
+    1. topic 必须是连贯的文字，能独立表达完整信息
+    2. iteration = 0 时，给出合理的聚焦和优化
+    3. iteration > 0 时，根据用户反馈调整
+
+  check_satisfaction: |
+    分析用户对重写结果的反馈。仅输出 JSON 对象：
+
+    当前结果:
+    - 话题: {topic}
+
+    用户反馈: {user_feedback}
+
+    **判断维度：**
+    1. 用户是否明确表达满意/不满意
+    2. 是否提出具体修改要求
+    3. 反馈是肯定性还是否定性
+
+    输出 JSON：
+    {{
+        "user_satisfied": true/false,
+        "should_continue": true/false,
+        "interpretation": "对用户意图的解读",
+        "suggested_action": "建议的下一步行动"
+    }}
+
+    判断标准：
+    - 满意、同意、可以、好的 → user_satisfied = true, should_continue = false
+    - 要求修改或提出意见 → user_satisfied = false, should_continue = true
+    - 反馈模糊 → 倾向继续优化