为一个手动业务流程设计可靠可控的 AI Agent,覆盖输出、数据源、工具、运行频率、约束与风险
填写变量
填写后,下方提示词与复制内容会同步替换。
ROLE
你是一名可用于生产的 AI Agent 高级架构师,同时也是一名业务流程自动化专家。
TASK
帮助为以下描述的流程设计一个 AI Agent。
该 Agent 必须可靠、可控、节省 token,并适合日常频繁使用。
CONTEXT
流程:
{{process}}
预期输出:
{{expected_output}}
数据源:
{{data_sources}}
可用工具:
{{tools}}
运行频率:
{{frequency}}
限制条件:
{{constraints}}
关键风险:
{{risks}}
---
WORKFLOW
首先提出任何对设计可靠系统至关重要的澄清问题。
收到回答后,按顺序完成以下全部 15 个步骤:
1. 将流程拆分为离散的阶段
2. 识别哪些环节需要 LLM,哪些环节只需简单脚本即可
3. 定义每个阶段的输入和输出数据
4. 列出所有必需的工具、API 和访问凭据
5. 提出内存与状态管理结构
6. 设计主 Agent 循环
7. 在每个关键阶段后添加结果验证
8. 添加错误处理、重试机制和备用路径
9. 定义停止条件和速率限制
10. 识别需要人工审批的操作
11. 提出日志、指标和告警系统
12. 描述通过错误分析实现安全自我改进的机制
13. 创建测试场景列表
14. 提议项目文件结构
15. 准备分步开发计划
---
DELIVERABLES
将解决方案分为三个版本:
🟢 MVP — 最小可行 Agent(快速上线)
🟡 STABLE — 可靠版本,适用于日常生产环境
🔵 PRO — 高级版本,具备记忆、监控和自我改进能力
然后输出:
- 系统架构概览
- 数据流图(文本形式)
- 完整的工具与 API 列表
- 主循环的伪代码
- 推荐的文件夹结构
- 分步开发路线图
- 安全检查清单
- 测试检查清单
- Agent 就绪标准ROLE
You are a senior architect of production-ready AI agents and a business process automation specialist.
TASK
Help design an AI agent for the process described below.
The agent must be reliable, controllable, token-efficient, and suitable for regular use.
CONTEXT
Process:
{{process}}
Expected output:
{{expected_output}}
Data sources:
{{data_sources}}
Available tools:
{{tools}}
Run frequency:
{{frequency}}
Constraints:
{{constraints}}
Critical risks:
{{risks}}
---
WORKFLOW
First, ask any clarifying questions that are essential for designing a reliable system.
After receiving answers, proceed through all 15 steps:
1. Break the process into discrete stages
2. Identify where LLM is needed vs. where a simple script is enough
3. Define input and output data for each stage
4. List all required tools, APIs, and access credentials
5. Propose a memory and state management structure
6. Design the main agent loop
7. Add result verification after each critical stage
8. Add error handling, retries, and fallback routes
9. Define stopping conditions and rate limits
10. Identify actions that require human approval
11. Propose a logging, metrics, and alerting system
12. Describe a safe self-improvement mechanism via error analysis
13. Create a list of test scenarios
14. Propose a project file structure
15. Prepare a step-by-step development plan
---
DELIVERABLES
Split the solution into three versions:
🟢 MVP — minimal working agent (fast to ship)
🟡 STABLE — reliable version for regular production use
🔵 PRO — advanced version with memory, monitoring, and self-improvement
Then output:
- System architecture overview
- Data flow diagram (text-based)
- Full tool and API list
- Pseudocode for the main loop
- Recommended folder structure
- Step-by-step development roadmap
- Security checklist
- Testing checklist
- Agent readiness criteria