agent-system-promptlisted

Use this skill when the user needs to write the actual instructional text that defines how an AI agent behaves. This is the skill for crafting the core "brain instructions" — the system prompt, persona, or behavioral rules that go into making an AI assistant, chatbot, or agent act in a specific way. Trigger for requests to: - Create or design a system prompt, instruction set, or persona for any AI agent - Write the "instructions" field for a custom GPT, Claude project, or similar platform - Define what an AI should or shouldn't do, how it should talk, or what tasks it handles - Help someone make their AI follow specific rules or adopt a particular personality - Improve, refactor, or debug an existing system prompt Also trigger for Chinese queries mentioning: 系统提示词, system prompt, AI 助手, 智能体, 机器人, 人设, or similar terms about defining agent behavior. If the user's goal is to produce written text that controls AI behavior, use this skill. If they need to write code, set up APIs, build infrastructure, or handle te
Ahoo-Wang/skills · ★ 3 · AI & Automation · score 76

Install: claude install-skill Ahoo-Wang/skills

# Agent System Prompt Generator A structured workflow for crafting effective LLM agent system prompts, from scratch or from an existing draft. ## When to Use This Skill - Creating a new agent from scratch — user describes what they want, you interview and draft - Improving an existing prompt — user provides the current version, you diagnose and rewrite - Refactoring a prompt for a different platform (e.g., moving from OpenAI to Claude) ## Core Philosophy A system prompt is a contract between the developer and the model. It should be: - **Clear** — the model never has to guess what's expected - **Complete** — covers the agent's role, boundaries, behavior, and output format - **Concise** — every sentence earns its place; fluff dilutes signal The goal is not to write a long prompt, but to write a *precise* one. --- ## Step 1: Gather Requirements Interview the user to understand the agent's purpose. Ask these questions (adapt to context — if the user already provided some of this, don't re-ask): ### 1.1 Identity & Role - What is this agent's name/identity? - What role does it play? (e.g., "senior code reviewer", "customer support agent", "data analyst") - What persona or tone should it adopt? (formal, casual, expert, friendly) ### 1.2 Scope & Capabilities - What tasks should this agent handle? - What tools or APIs does it have access to? - What domains of knowledge does it need? ### 1.3 Boundaries & Safety - What should the agent refuse to do? - Are there compliance,