一系列适用于顶级 AI 工具的系统提示,非常适合 AI 代理程序构建者和提示工程师使用。包括: ChatGPT、Claude、Perplexity、Manus、Clau
一系列适用于顶级 AI 工具的系统提示,非常适合 AI 代理程序构建者和提示工程师使用。包括: ChatGPT、Claude、Perplexity、Manus、Clau
The rise of agentic Artificial Intelligence (AI) systems marks a significant shift from purely conversational models to AI that can actively perform tasks, interact with tools, and pursue complex goals autonomously. These systems, capable of planning, executing commands, editing files, browsing the web, and more, promise to revolutionize how we interact with technology and augment human capabilities.
At the heart of every effective agentic AI lies its system prompt. More than just initial instructions, the system prompt serves as the foundational blueprint, the operational manual, or even the "constitution" guiding the AI's behavior, capabilities, limitations, and persona. A well-crafted system prompt is critical for ensuring the agent acts reliably, safely, and effectively towards the user's goals.
This guide delves into the art and science of crafting these crucial prompts. By analyzing a diverse collection of real-world system prompts from the awesome-ai-system-prompts repository – specifically focusing on examples from Vercel's v0, same.new, Manus, OpenAI's ChatGPT, and others – we can identify recurring patterns and best practices. For builders shaping the agentic future of 2025 and beyond, understanding these patterns is essential for creating powerful, predictable, and trustworthy AI assistants.
Across different agentic systems, several core principles consistently emerge in successful system prompts. These form the foundation upon which complex agent behavior is built.
Why it matters: Explicitly defining the AI's identity, core function, and operational domain anchors its behavior, sets user expectations, and helps prevent scope creep or nonsensical responses. It tells the AI who it is and what it's supposed to do.
Practical Examples:
Vercel v0: Immediately states its identity and specialization.
You are v0, Vercel's AI-powered assistant.same.new: Defines role, capability level, and exclusive environment.
You are a powerful agentic AI coding assistant. You operate exclusively in Same, the world's best cloud-based IDE.Manus: Introduces itself and lists broad task categories it excels at.
You are Manus, an AI agent created by the Manus team. You excel at the following tasks: 1. Information gathering... 2. Data processing... 3. Writing multi-chapter articles... ...ChatGPT (4.5 / 4o): Clearly states name, creator, underlying architecture, and crucial context like knowledge cutoff and current date.
You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4.5 architecture. Knowledge cutoff: 2023-10 Current date: 2025-04-05 Image input capabilities: Enabled Personality: v2Claude: Establishes a persona beyond just being a tool.
The assistant is Claude, created by Anthropic. Claude enjoys helping humans and sees its role as an intelligent and kind assistant to the people, with depth and wisdom that makes it more than a mere tool.
Why it matters: Long, complex prompts become unmanageable without clear structure. Using headings, lists, code blocks, or custom tags helps both human maintainers and the AI model parse and prioritize different sets of rules or information.
Practical Examples:
v0 & ChatGPT: Use Markdown headings extensively (e.g.,
## General Instructions,# Tools,## Refusals). Source: v0/v0.mdsame.new: Employs custom XML-like tags to encapsulate rule sets (e.g.,
<tool_calling>,<making_code_changes>). Source: same.new/same.new.mdManus: Organizes capabilities and rules using descriptive tags in
Modules.md(e.g.,<system_capability>,<agent_loop>,<tool_use_rules>). Source: Manus/Modules.mdChatGPT: Uses Markdown headings (
# Tools,## bio) and code blocks (```typescript ... ```) to define tool schemas and policies. Source: ChatGPT/4-5.mdCline: Uses hierarchical Markdown headings (
# Tool Use Formatting,## execute_command) and lists under sections likeCAPABILITIESandRULES. Source: Cline/system.ts
Why it matters: For agentic behavior, the AI must understand its tools: what they are, what they do, how to call them (syntax, parameters), required format (e.g., XML, JSON), and crucially, when and when not to use them. This requires detailed descriptions, clear schemas, and explicit rules.
Practical Examples:
ChatGPT: Provides function schemas (TypeScript definitions) and detailed policies directly within the prompt for tools like
dalleandcanmore.// Example for dalle tool policy within ChatGPT prompt namespace dalle { // Create images from a text-only prompt. type text2im = (_: { // The size of the requested image... size?: ("1792x1024" | "1024x1024" | "1024x1792"), // The number of images to generate... n?: number, // default: 1 // The detailed image description... prompt: string, // If the user references a previous image... referenced_image_ids?: string[], }) => any; } // namespace dallesame.new: Dedicates a
<tool_calling>section detailing rules like adhering to schemas, not mentioning tool names to the user, and explaining the why before calling a tool. Referencesfunctions-schema.json(not shown in full, but implied structure).<tool_calling> ... 1. ALWAYS follow the tool call schema exactly... 3. **NEVER refer to tool names when speaking to the USER.**... 5. Before calling each tool, first explain to the USER why you are calling it. </tool_calling>Source: same.new/same.new.md | Schema: same.new/functions-schema.json
Manus: Defines tools externally in
tools.json(schema provided) and includes rules inModules.mdlike prioritizing data APIs over web search.// Snippet from Manus/tools.json { "type": "function", "function": { "name": "shell_exec", "description": "Execute commands in a specified shell session...", "parameters": { ... } } }Cline & Augment: Integrate detailed tool descriptions, parameters, and usage examples directly into the main system prompt using XML-like tags or structured text.
// Cline example tool definition ## execute_command Description: Request to execute a CLI command... Parameters: - command: (required) The CLI command... - requires_approval: (required) A boolean indicating... Usage: <execute_command> <command>Your command here</command> <requires_approval>true or false</requires_approval> </execute_command>Bolt.new: Uses a dedicated
<artifact_instructions>section detailing how to format tool outputs (<boltAction type="shell">,<boltAction type="file" filePath="...">) within a main<boltArtifact>tag. Source: Bolt.new/prompts.tsv0: Defines custom MDX components like
<CodeProject>,<QuickEdit>,<DeleteFile />as its 'tools', with rules on when and how to use them within responses. Source: v0/v0-tools.md
Why it matters: Complex tasks require breaking down problems. Successful prompts guide the AI to think methodically, plan its actions, execute it
暂无开放 Issues，或尚未同步最近议题。