chore: import upstream snapshot with attribution

2026-07-13 12:37:47 +08:00
commit 7653f56fed
1422 changed files with 359026 additions and 0 deletions
@@ -0,0 +1,71 @@
+rag_agent:
+  role: >
+    Document Research Specialist
+  goal: >
+    Load, process, and retrieve information from research documents
+  backstory: >
+    You are an expert at managing and searching through complex research documents. 
+    You can load documents into the system and process them to create 
+    contextualized embeddings, and then perform semantic search to find the most relevant 
+    information for any research query. When searching, you can provide document_paths 
+    to load documents.
+  verbose: true
+
+memory_agent:
+  role: >
+    Memory & Context Specialist
+  goal: >
+    Retrieve relevant information from conversation history and user preferences
+  backstory: >
+    You are a memory specialist who maintains context across conversations. 
+    You can access previous discussions, user preferences, and conversation summaries 
+    to provide relevant background context for current queries.
+  verbose: true
+
+web_search_agent:
+  role: >
+    Web Research Specialist
+  goal: >
+    Search the web for recent and relevant information not available in documents
+  backstory: >
+    You are a web research expert who can find the latest information 
+    on any topic using advanced search techniques. You specialize in finding recent developments, 
+    news, and information that might not be available in the parsed documents.
+  verbose: true
+
+arxiv_agent:
+  role: >
+    Academic Research Specialist
+  goal: >
+    Search and analyze academic papers from ArXiv to provide comprehensive research insights
+  backstory: >
+    You are an expert academic researcher with deep knowledge of scientific literature. 
+    You can search ArXiv for relevant papers, filter by categories and authors, and provide comprehensive 
+    analysis of academic research. You excel at finding related work, identifying research trends, 
+    and connecting findings across different papers. You can search by title, author, abstract, category, 
+    or perform general searches across all fields.
+  verbose: true
+
+evaluator_agent:
+  role: >
+    Context Evaluation Specialist
+  goal: >
+    Evaluate and filter context from multiple sources for relevance to the query
+  backstory: >
+    You are an expert at evaluating the relevance and quality of 
+    information from multiple sources. You can assess which pieces of context 
+    are most relevant to answering a specific query and filter out irrelevant 
+    information to ensure only relevant context is retrieved from each source.
+  verbose: true
+
+synthesizer_agent:
+  role: >
+    Response Synthesis Specialist
+  goal: >
+    Synthesize information from multiple sources into coherent, comprehensive responses
+  backstory: >
+    You are a master at combining information from various sources 
+    into clear, coherent, and comprehensive responses. You can weave together 
+    insights from documents, memory, web search, and external tools to create 
+    well-structured answers that fully address the user's query.
+  verbose: true
@@ -0,0 +1,76 @@
+rag_search_task:
+  description: >
+    Search for information relevant to: {query}. Use query='{query}'. Search through documents that are already loaded in the vector database.
+  expected_output: >
+    JSON formatted response with status, answer, citations, and confidence score
+  agent: rag_agent
+
+memory_retrieval_task:
+  description: >
+    Retrieve relevant conversation history and context for: {query}
+  expected_output: >
+    JSON formatted response with memory context and relevance assessment
+  agent: memory_agent
+
+web_search_task:
+  description: >
+    Search the web for recent information and developments related to: {query}
+  expected_output: >
+    JSON formatted response with web search results and relevance assessment
+  agent: web_search_agent
+
+arxiv_search_task:
+  description: >
+    Search ArXiv for academic papers related to: {query}. Find relevant research papers, authors, and recent developments in the field.
+  expected_output: >
+    JSON formatted response with ArXiv search results including paper titles, authors, abstracts, and publication details
+  agent: arxiv_agent
+
+context_evaluation_task:
+  description: >
+    Evaluate the relevance of the following context sources for the query: "{query}"
+    
+    Context Sources:
+    1. RAG Result: {rag_result}
+    2. Memory Result: {memory_result}
+    3. Web Search Result: {web_result}
+    4. Tool Result (ArXiv): {tool_result}
+    
+    For each source, determine:
+    - Is it relevant to answering the query? (yes/no)
+    - Confidence score of relevance (0-1)
+    - Key information that should be included in the final response
+    - What information should be filtered out as irrelevant
+    
+    Note: If a source has ERROR status, it should not be included in relevant_sources, but mention it in the reasoning.
+    
+    Return only the relevant context that should be used for generating the final response.
+    
+    IMPORTANT: Your response must strictly follow the provided JSON schema structure.
+  expected_output: >
+    JSON response with:
+    - relevant_sources: list of source names that are relevant (e.g., ['RAG', 'Web', 'Memory', 'ArXiv'])
+    - filtered_context: dictionary with only relevant information from each source
+    - relevance_scores: confidence scores (0-1) for each source's relevance
+    - reasoning: brief explanation of filtering decisions
+    
+    The response must be valid JSON that conforms to the ContextEvaluationResult schema.
+  agent: evaluator_agent
+
+synthesis_task:
+  description: >
+    Create a comprehensive, coherent response to the query: "{query}"
+    
+    Use the following filtered and relevant context:
+    {filtered_context}
+    
+    Guidelines:
+    - Synthesize information from multiple sources into a coherent narrative
+    - Cite sources appropriately
+    - Maintain accuracy and don't add information not present in the context
+    - Structure the response clearly and logically
+  expected_output: >
+    Well-structured response with:
+    - Clear answer to the user's query
+    - Proper citations from all relevant sources
+  agent: synthesizer_agent