PromptBase
PromptBase SelectGet Select
Close icon
General
Home
Marketplace
Create
Hire
Login
Chat
Sell
Explore

Prompt Details

Model
Chat - (gpt-6.1-sol)
Token size
726
Example input
[BUSINESS_PROCESS_TO_AUTOMATE]: B2B SaaS Customer Support Ticket Triage, Bug Reproduction & Billing Refund Resolution [TOOLS_AND_STACK]: LangGraph (Python), Zendesk API, Stripe API, PostgreSQL Vector DB, Slack Webhook [GUARDRAILS_AND_GOALS]: Human approval required for refunds over $50; zero hallucinated policy promises; under 12s response SLA
Example output
### MODULE 1: Multi-Agent Topology & State Graph 1. Orchestration Pattern: Supervisor-Worker with Conditional Router (LangGraph StateGraph). 2. Agent Roster: - Triage_Router_Agent (gpt-4o-mini): Classifies ticket intent, urgency (P0-P4), and masks PII. - RAG_Knowledge_Agent (gpt-4o): Queries PostgreSQL pgvector docs for verified bug reproduction steps. - Billing_Resolution_Agent (gpt-4o): Audits Stripe charges and refund policy (<14 days, <100 API calls). - QA_Synthesis_Agent (gpt-4o-mini): Verifies policy guardrails and posts final Zendesk reply. 3. Execution Flow: [Zendesk Webhook] -> [Triage_Router_Agent] -> (Technical: [RAG_Knowledge_Agent] | Billing: [Billing_Resolution_Agent]) -> (If Refund > $50: [Slack HITL Approval Gate]) -> [QA_Synthesis_Agent] -> [Zendesk Reply API] ### MODULE 2: Production System Prompts - Triage_Router_Agent: "Classify incoming Zendesk tickets into TECHNICAL_SUPPORT, BILLING_DISPUTE, or CRITICAL_OUTAGE. Redact all credit card numbers and API keys. Output strict JSON: {intent_category, urgency_score, sanitized_query}." - RAG_Knowledge_Agent: "Answer strictly using retrieved pgvector doc_ids. Never promise unreleased features. If cosine similarity < 0.78, set confidence_score < 0.80 and route to human Tier-2." - Billing_Resolution_Agent: "Verify Stripe charge age and usage. Never invoke issue_stripe_refund if refund_amount_usd > 50.00; instead set hitl_required: true." ### MODULE 3: Tool-Calling JSON Schema Contract { "name": "issue_stripe_refund", "description": "Executes autonomous Stripe refund up to $50.00 or stages higher amounts for human approval.", "parameters": { "type": "object", "properties": { "charge_id": { "type": "string" }, "refund_amount_cents": { "type": "integer", "maximum": 5000 }, "reason": { "type": "string", "enum": ["duplicate", "fraudulent", "requested_by_customer"] }, "idempotency_key": { "type": "string" } }, "required": ["charge_id", "refund_amount_cents", "reason", "idempotency_key"] } } ### MODULE 4: Shared State Schema & Handoff Protocol class SupportWorkflowState(TypedDict): zendesk_ticket_id: str intent_category: str confidence_score: float refund_amount_usd: float hitl_required: bool hitl_approved: bool final_customer_reply: str - Handoff Rules: Route to HITL_Interrupt_Node whenever refund_amount_usd > 50.00 or confidence_score < 0.80. ### MODULE 5: Error Recovery & Edge-Case Fallbacks - Schema Validation Error: Auto-retry (max 2x) feeding Pydantic ValidationError back to the agent. - Stripe/Zendesk API 429 Timeout: Exponential backoff (1s, 2s, 4s) with PostgreSQL checkpoint persistence. - Prompt Injection Defense: Read charge amounts strictly from Stripe API metadata, ignoring user-claimed amounts. ### MODULE 6: Human-in-the-Loop (HITL) Gate & Deployment Checklist 1. HITL Gate: Configure interrupt_before=["execute_high_value_refund"] in LangGraph with PostgresSaver; send interactive Approve/Reject Slack Block Kit alert to #billing-approvals. 2. Deployment: Enforce server-side assertion (amount_cents <= 5000 or hitl_approved == True) inside the FastAPI webhook worker to guarantee <12s SLA.
GPT

Multiagent Workflow Architect Pro

Add to Cart
PromptBase Select
Get this prompt + 9 more with Select$14 your first month, then $19/mo$14 first month, then $19/mo
View Select
Instant accessInstant access
Usage rightsCommercial use
Money-back guaranteeMoney‑back
By purchasing this prompt, you agree to our terms of service
GPT-6.1-SOL
Tested icon
Guide icon
4 examples icon
Free credits icon
Turn any complex business process into a production-ready Multi-Agent AI Automation Blueprint. Generates deterministic agent roles, system prompts, JSON tool-calling schemas, state-routing logic, error-handling fallbacks, and Human-in-the-Loop (HITL) safety guardrails compatible with LangGraph, CrewAI, n8n, Make, and OpenAI Agents SDK. Ideal for AI agencies, SaaS builders, and automation engineers.
...more
Added 1 week ago
Report
Browse Marketplace