Release v2.0.1

Develop

Author: WMC001

backend/agents/create_agent_info.py

@@ -14,6 +14,7 @@
     ElasticSearchService,
     get_vector_db_core,
     get_embedding_model,
+    get_rerank_model,
 )
 from services.remote_mcp_service import get_remote_mcp_server_list
 from services.memory_config_service import build_memory_context
@@ -296,8 +297,6 @@ async def create_agent_config(
     }
     system_prompt = Template(prompt_template["system_prompt"], undefined=StrictUndefined).render(render_kwargs)
 
-    _print_prompt_with_token_count(system_prompt, agent_id, "BEFORE_INJECTION")
-
     if agent_info.get("model_id") is not None:
         model_info = get_model_by_model_id(agent_info.get("model_id"))
         model_name = model_info["display_name"] if model_info is not None else "main_model"
@@ -350,11 +349,32 @@ async def create_tool_config_list(agent_id, tenant_id, user_id, version_no: int
                     tool_config.metadata = langchain_tool
                     break
 
-        # special logic for knowledge base search tool
+        # special logic for search tools that may use reranking models
         if tool_config.class_name == "KnowledgeBaseSearchTool":
-           tool_config.metadata = {
+            rerank = param_dict.get("rerank", False)
+            rerank_model_name = param_dict.get("rerank_model_name", "")
+            rerank_model = None
+            if rerank and rerank_model_name:
+                rerank_model = get_rerank_model(
+                    tenant_id=tenant_id, model_name=rerank_model_name
+                )
+
+            tool_config.metadata = {
                 "vdb_core": get_vector_db_core(),
                 "embedding_model": get_embedding_model(tenant_id=tenant_id),
+                "rerank_model": rerank_model,
+            }
+        elif tool_config.class_name in ["DifySearchTool", "DataMateSearchTool"]:
+            rerank = param_dict.get("rerank", False)
+            rerank_model_name = param_dict.get("rerank_model_name", "")
+            rerank_model = None
+            if rerank and rerank_model_name:
+                rerank_model = get_rerank_model(
+                    tenant_id=tenant_id, model_name=rerank_model_name
+                )
+
+            tool_config.metadata = {
+                "rerank_model": rerank_model,
             }
         elif tool_config.class_name == "AnalyzeTextFileTool":
             tool_config.metadata = {
@@ -430,25 +450,9 @@ async def prepare_prompt_templates(
     prompt_templates = get_agent_prompt_template(is_manager, language)
     prompt_templates["system_prompt"] = system_prompt
 
-    # Print final prompt with all injections
-    _print_prompt_with_token_count(prompt_templates["system_prompt"], agent_id, "FINAL_PROMPT")
-
     return prompt_templates
 
 
-def _print_prompt_with_token_count(prompt: str, agent_id: int = None, stage: str = "PROMPT"):
-    """Print prompt content and estimate token count using tiktoken."""
-    try:
-        import tiktoken
-        encoding = tiktoken.get_encoding("cl100k_base")
-        token_count = len(encoding.encode(prompt))
-        logger.info(f"[Skill Debug][{stage}] Agent {agent_id} token count: {token_count}")
-        logger.info(f"[Skill Debug][{stage}] Agent {agent_id} prompt:\n{prompt}")
-    except Exception as e:
-        logger.warning(f"[Skill Debug][{stage}] Failed to count tokens: {e}")
-        logger.info(f"[Skill Debug][{stage}] Agent {agent_id} prompt:\n{prompt}")
-
-
 async def join_minio_file_description_to_query(minio_files, query):
     final_query = query
     if minio_files and isinstance(minio_files, list):
@@ -527,7 +531,7 @@ async def create_agent_run_info(
     remote_mcp_list = await get_remote_mcp_server_list(tenant_id=tenant_id, is_need_auth=True)
     default_mcp_url = urljoin(LOCAL_MCP_SERVER, "sse")
     remote_mcp_list.append({
-        "remote_mcp_server_name": "nexent",
+        "remote_mcp_server_name": "outer-apis",
         "remote_mcp_server": default_mcp_url,
         "status": True,
         "authorization_token": None

backend/agents/skill_creation_agent.py

@@ -0,0 +1,122 @@
+"""Skill creation agent module for interactive skill generation."""
+
+import logging
+import threading
+from typing import List
+
+from nexent.core.agents.agent_model import AgentConfig, AgentRunInfo, ModelConfig, ToolConfig
+from nexent.core.agents.run_agent import agent_run_thread
+from nexent.core.utils.observer import MessageObserver
+
+logger = logging.getLogger("skill_creation_agent")
+
+
+def create_skill_creation_agent_config(
+    system_prompt: str,
+    model_config_list: List[ModelConfig],
+    local_skills_dir: str = ""
+) -> AgentConfig:
+    """
+    Create agent config for skill creation with builtin tools.
+
+    Args:
+        system_prompt: Custom system prompt to replace smolagent defaults
+        model_config_list: List of model configurations
+
+    Returns:
+        AgentConfig configured for skill creation
+    """
+    if not model_config_list:
+        raise ValueError("model_config_list cannot be empty")
+
+    first_model = model_config_list[0]
+
+    prompt_templates = {
+        "system_prompt": system_prompt,
+        "managed_agent": {
+            "task": "{task}",
+            "report": "## {name} Report\n\n{final_answer}"
+        },
+        "planning": {
+            "initial_plan": "",
+            "update_plan_pre_messages": "",
+            "update_plan_post_messages": ""
+        },
+        "final_answer": {
+            "pre_messages": "",
+            "post_messages": ""
+        }
+    }
+
+    return AgentConfig(
+        name="__skill_creator__",
+        description="Internal skill creator agent",
+        prompt_templates=prompt_templates,
+        tools=[],
+        max_steps=5,
+        model_name=first_model.cite_name
+    )
+
+
+def run_skill_creation_agent(
+    query: str,
+    agent_config: AgentConfig,
+    model_config_list: List[ModelConfig],
+    observer: MessageObserver,
+    stop_event: threading.Event,
+) -> None:
+    """
+    Run the skill creator agent synchronously.
+
+    Args:
+        query: User query for the agent
+        agent_config: Pre-configured agent config
+        model_config_list: List of model configurations
+        observer: Message observer for capturing agent output
+        stop_event: Threading event for cancellation
+    """
+    agent_run_info = AgentRunInfo(
+        query=query,
+        model_config_list=model_config_list,
+        observer=observer,
+        agent_config=agent_config,
+        stop_event=stop_event
+    )
+
+    agent_run_thread(agent_run_info)
+
+
+def create_simple_skill_from_request(
+    system_prompt: str,
+    user_prompt: str,
+    model_config_list: List[ModelConfig],
+    observer: MessageObserver,
+    stop_event: threading.Event,
+    local_skills_dir: str = ""
+) -> None:
+    """
+    Run skill creation agent to create a skill interactively.
+
+    The agent will write the skill content to tmp.md in local_skills_dir.
+    Frontend should read tmp.md after agent completes to get the skill content.
+
+    Args:
+        system_prompt: System prompt with skill creation instructions
+        user_prompt: User's skill description request
+        model_config_list: List of model configurations
+        observer: Message observer for capturing agent output
+        stop_event: Threading event for cancellation
+        local_skills_dir: Path to local skills directory for file operations
+    """
+    agent_config = create_skill_creation_agent_config(
+        system_prompt=system_prompt,
+        model_config_list=model_config_list,
+        local_skills_dir=local_skills_dir
+    )
+
+    thread_agent = threading.Thread(
+        target=run_skill_creation_agent,
+        args=(user_prompt, agent_config, model_config_list, observer, stop_event)
+    )
+    thread_agent.start()
+    thread_agent.join()