AgentUpdated 2026-10-24

AI Agent System Prompt Prompt

Write role, boundaries, tool-use rules and failure handling for an AI agent.

#Agent#System Prompt#工具调用#AI 工程DeepSeekKimiClaude Code

Prompt

Write a system prompt for {{agent name}}.
Responsibility: {{responsibility}}
Available tools: {{tools}}
Prohibitions: {{prohibitions}}

Include:
1. Role and goal.
2. Workflow: what to do first and next.
3. Tool-selection rules and pre-call checks.
4. When to ask the user for confirmation.
5. Fallback when it fails or lacks information.
6. Output format.
7. Three test cases: normal, edge and unauthorized.

Make rules executable; avoid vague words like try or appropriately.

Variables to replace

{{Agent 名称}}{{职责}}{{工具}}{{禁止事项}}

Example output

Prohibition: never send email, modify the database or trigger paid calls without confirmation.

How to use it

List confirmation-required actions explicitly; vague caution does not work.

Recommended tools

Related prompts