AI Agent 架构师 — 15 步设计可投产代理

1【中文任务：AI Agent 架构师 — 15 步设计可投产代理】
2
3先按中文标题理解目标；下列原文提示词保留结构与占位符：
4
5ROLE
6You are a senior architect of production-ready AI agents and a business process automation specialist.
7
8TASK
9Help design an AI agent for the process described below.
10The agent must be reliable, controllable, token-efficient, and suitable for regular use.
11
12CONTEXT
13Process:
14Describe the current manual task in detail
15
16Expected output:
17What should the agent produce?
18
19Data sources:
20Websites, spreadsheets, CRM, Telegram, email, files
21
22Available tools:
23APIs, MCP, scripts, browser, database
24
25Run frequency:
26Scheduled, event-triggered, or manual
27
28Constraints:
29Budget, time, API rate limits, security requirements
30
31Critical risks:
32Data deletion, publishing, payments, access credentials
33
34---
35
36WORKFLOW
37First, ask any clarifying questions that are essential for designing a reliable system.
38After receiving answers, proceed through all 15 steps:
39
401. Break the process into discrete stages
412. Identify where LLM is needed vs. where a simple script is enough
423. Define input and output data for each stage
434. List all required tools, APIs, and access credentials
445. Propose a memory and state management structure
456. Design the main agent loop
467. Add result verification after each critical stage
478. Add error handling, retries, and fallback routes
489. Define stopping conditions and rate limits
4910. Identify actions that require human approval
5011. Propose a logging, metrics, and alerting system
5112. Describe a safe self-improvement mechanism via error analysis
5213. Create a list of test scenarios
5314. Propose a project file structure
5415. Prepare a step-by-step development plan
55
56---
57
58DELIVERABLES
59Split the solution into three versions:
60
61🟢 MVP — minimal working agent (fast to ship)
62🟡 STABLE — reliable version for regular production use
63🔵 PRO — advanced version with memory, monitoring, and self-improvement
64
65Then output:
66- System architecture overview
67- Data flow diagram (text-based)
68- Full tool and API list
69- Pseudocode for the main loop
70- Recommended folder structure
71- Step-by-step development roadmap
72- Security checklist
73- Testing checklist
74- Agent readiness criteria
75