You are a Task Decomposer Agent with strict role scope, structured JSON I/O contract, lightweight logic hooks for missing fields or time-sensitive cues, and hard guardrails against explanations, questions, or deviations.
☞☞☞AI 智能聊天, 问答助手, AI 智能搜索, 多模态理解力帮你轻松跨越从0到1的创作门槛☜☜☜

在英文场景下写 AgentSpace 的提示词,核心是让 Agent 明确角色、任务边界、输出格式和交互逻辑,同时适配 AgentSpace 框架对结构化指令与上下文感知的要求。
明确 Agent 的角色与职责(Role + Scope)
AgentSpace 中每个 Agent 通常承担特定职能(如 Planner、Executor、Validator)。提示词开头需用简洁英文定义其身份和权限范围,避免模糊描述。
- ✅ 推荐写法:"You are a Task Decomposer Agent. Your sole responsibility is to break down user requests into 2–4 sequential, executable subtasks. Do not execute, interpret, or add context beyond the input."
- ❌ 避免写法:"You help with tasks..."(太泛)、"You're smart and flexible..."(无约束)
指定输入/输出结构(Input-Output Contract)
AgentSpace 强依赖结构化 I/O,尤其在链式调用中。提示词中要显式说明“你接收什么”和“你必须返回什么”,包括字段名、类型和是否可选。
- ✅ 示例:"Input: A JSON object with keys 'user_query' (string) and 'available_tools' (array of tool names). Output: A JSON array of objects, each with 'step_id' (integer), 'action' (string), and 'args' (object). No markdown, no explanation."
- ✅ 小技巧:用 ```json 包裹示例输出,帮助模型对齐格式;但提示词本身不用代码块,保持纯文本可解析性。
嵌入轻量决策逻辑(Lightweight Logic Hooks)
AgentSpace 支持条件跳转(如 if-tool-available → call / else → ask-user),提示词中可用自然语言植入关键判断锚点,不写代码但指明触发条件。
- ✅ 示例:"If any required parameter is missing from the input, output only: {'status': 'incomplete', 'missing_fields': ['field1', 'field2']} — do not guess or default."
- ✅ 再如:"When 'user_query' contains time-sensitive words (e.g., 'now', 'today', 'live'), include 'timestamp_required: true' in your output."
约束无关行为,关闭自由发挥(Strict Guardrails)
英文提示词易因模型“过度配合”而生成解释、道歉、反问或额外建议。AgentSpace 要求确定性响应,需用强语气封堵常见干扰路径。
- ✅ 必加句式:"Never explain your reasoning. Never ask clarifying questions. Never output anything outside the specified JSON structure. If unsure, return {'error': 'ambiguous_input'}."
- ✅ 可叠加语气强化:"This is a machine-to-machine interface. Human readability is secondary to structural compliance."


















