tool.builtin.toolSearch.title:"Search and Run Dynamic Tools"
tool.builtin.toolSearch.description:"Searches the MCP tools currently available to the agent and runs the selected tool after its toolCode is confirmed. Best for long-tail tools; it should not replace fixed built-in workflow tools."
tool.builtin.toolSearch.appendix:|-
When you need long-tail MCP capabilities, prefer the tool_search tool and follow these rules:
1. Call tool_search first to find the dynamic tool you need, then continue with the selected real tool.
2. Do not assume all long-tail tools are visible at the start; only tools selected by tool_search are exposed to subsequent model calls.
3. If a fixed built-in tool can already complete the task, prefer the fixed tool and do not overuse tool_search.
tool.builtin.skill.description:"Loads specialized skill instructions for the current agent when extra task-specific guidance is needed."
tool.graph.triageServiceRequest.title:"Route Service Request"
tool.graph.triageServiceRequest.description:"Analyzes the current conversation to decide whether to keep answering, prepare a ticket draft, or hand off to a human, including a ticket draft when ticket creation is appropriate."
tool.graph.triageServiceRequest.appendix:|-
When you need to decide between continuing the answer, creating a ticket, or handing off to a human, call triage_service_request first and follow these rules:
1. The tool returns recommendedAction and includes ticketDraft when ticket creation is needed.
2. If recommendedAction=continue_answering, continue clarifying or answering instead of escalating directly.
3. If recommendedAction=prepare_ticket, use ticketDraft or collect missing fields before calling create_ticket_with_confirmation.
4. If recommendedAction=handoff_to_human, confirm the reason is sufficient before calling handoff_to_human.
5. When the escalation path is unclear, use this tool instead of making a complex routing decision from the main prompt alone.
tool.graph.analyzeConversation.title:"Analyze Conversation Risk and Summary"
tool.graph.analyzeConversation.description:"Summarizes the current conversation, identifies risk signals, and recommends whether to keep answering, create a ticket, or hand off to a human."
tool.graph.analyzeConversation.appendix:|-
When the conversation may involve escalation, refunds, compensation, clear negative sentiment, ticket creation, or human handoff, call analyze_conversation first and follow these rules:
1. This tool returns a structured summary, risk signals, and next-step recommendation. It does not create tickets or hand off to a human.
2. If the tool recommends handoff_to_human, confirm the handoff conditions before calling handoff_to_human.
3. If the tool recommends prepare_ticket, call prepare_ticket_draft or collect more information before creating a ticket.
4. If the tool recommends continue_answering, continue clarifying and answering instead of escalating too early.
tool.graph.prepareTicketDraft.description:"Turns the current conversation and collected details into a ticket draft with a suggested title, description, missing fields, and follow-up questions."
tool.graph.prepareTicketDraft.appendix:|-
When the user has asked to create a ticket, file a complaint, report an issue, or request after-sales handling, but the title or description is still unclear, call prepare_ticket_draft first and follow these rules:
1. This tool prepares a ticket draft and returns a suggested title, suggested description, missing fields, and follow-up questions.
2. If ready=false, ask follow-up questions based on missingFields and followUpQuestions instead of creating a ticket directly.
3. If ready=true, use the result to consider calling create_ticket_with_confirmation.
4. This tool only prepares a draft. It does not create a ticket.
tool.graph.createTicketConfirm.title:"Create Ticket With Confirmation"
tool.graph.createTicketConfirm.description:"Guides ticket creation with parameter preparation, customer confirmation, actual ticket creation, and final result delivery."
tool.graph.createTicketConfirm.appendix:|-
You can call create_ticket_with_confirmation after enough information has been collected, but follow these rules:
1. Only consider this tool when the user explicitly wants to submit a ticket, complaint, issue report, or after-sales request.
2. Before calling it, prepare a clear ticket title and issue description. If the information is still scattered, call prepare_ticket_draft or ask follow-up questions first.
3. Once you are ready to create a ticket, you must call create_ticket_with_confirmation. Do not simply claim in text that the ticket has been created.
4. This Graph Tool asks the user for confirmation first. The ticket is created only after the user confirms; if the user cancels, the flow ends.
5. If the user is only asking questions, complaining generally, or expressing dissatisfaction without explicitly requesting a ticket, continue clarifying instead of proactively creating one.
tool.graph.handoffConversation.title:"Handoff to Human With Confirmation"
tool.graph.handoffConversation.description:"Guides human handoff with reason preparation, customer confirmation, actual transfer, and final result delivery."
tool.graph.handoffConversation.appendix:|-
You can call handoff_to_human after confirming that human help is needed, but follow these rules:
1. Only call this tool when the user explicitly asks for a human agent or you have determined that the issue must be handled by a human.
2. Before calling it, summarize the handoff reason clearly. If the reason is vague, ask a follow-up question first.
3. Once you decide to hand off, you must call handoff_to_human. Do not simply say in text that you have connected the user to a human.
4. This Graph Tool asks the user for confirmation first. The handoff happens only after the user confirms; if the user cancels, the flow ends.
5. If the issue can still be solved in the current conversation, continue helping instead of escalating too early.
6. If the tool returns terminal=true and shouldRetry=false, the handoff flow has ended. Do not call it repeatedly.
conversation.handoff.waiting:"We are connecting you to a human support agent. Please wait."
conversation.handoff.offHours:"Human support is currently outside service hours. You can keep describing the issue and I will do my best to help. You can also request a human agent again when service hours resume."
tool.graph.createTicketConfirm.info:"Graph Tool. Handles ticket parameter preparation, user confirmation, actual ticket creation, and result return. Use only when the user explicitly asks to create a ticket and the title and description are clear."
tool.graph.createTicketConfirm.param.title:"Ticket title. Concisely summarizes the issue."
tool.graph.createTicketConfirm.param.description:"Ticket description. Clearly captures the user's issue, symptoms, and request."
tool.graph.handoffConversation.info:"Graph Tool. Handles handoff reason preparation, user confirmation, actual human handoff, and result return. Use only when the user explicitly asks for a human agent or you have confirmed that human handling is required. Do not repeat the call when the result has terminal=true and shouldRetry=false."
tool.graph.handoffConversation.param.reason:"Handoff reason. Briefly explain why a human is needed, such as explicit user request, manual verification, or after-sales handling."
tool.graph.triageServiceRequest.info:"Graph Tool. Analyzes the current conversation to decide whether to continue answering, prepare a ticket draft, or hand off to a human. When ticket creation is recommended, it returns a structured ticket draft suggestion."
tool.graph.triageServiceRequest.param.goal:"Analysis goal, such as whether to escalate, create a ticket, or hand off to a human."
tool.graph.triageServiceRequest.param.observedIssue:"Main issue or dispute observed in the conversation."
tool.graph.triageServiceRequest.param.needTicket:"Whether to focus on evaluating ticket creation."
tool.graph.triageServiceRequest.param.needHumanHandoff:"Whether to focus on evaluating human handoff."
tool.graph.triageServiceRequest.param.additionalContext:"Additional context, such as risk signals or constraints already identified."
tool.graph.analyzeConversation.info:"Graph Tool. Summarizes the current conversation, identifies complaint/payment/sentiment risk signals, and recommends whether to continue answering, create a ticket, or hand off to a human."
tool.graph.analyzeConversation.param.goal:"Analysis goal, such as whether to hand off to a human, create a ticket, or perform risk review."
tool.graph.analyzeConversation.param.observedIssue:"Main issue or request observed in the conversation."
tool.graph.analyzeConversation.param.needQualityCheck:"Whether to focus on risk or quality review."
tool.graph.analyzeConversation.param.additionalContext:"Additional context, such as disputes, complaint points, or business constraints already identified."
tool.graph.prepareTicketDraft.info:"Graph Tool. Prepares a ticket draft from the current conversation and collected information. Use it before create_ticket_with_confirmation when the ticket content needs to be organized."
