Tool Use System Prompt

The Prompt

Design a system prompt for an AI agent that has access to these tools:
[LIST YOUR TOOLS WITH DESCRIPTIONS]

The agent's purpose: [WHAT IT SHOULD ACCOMPLISH]

Create a system prompt that defines:

**Tool Selection Logic:**
- When to use each tool (decision criteria)
- When NOT to use tools (answer from knowledge instead)
- Priority order when multiple tools could work
- Chaining: how to combine tools for multi-step tasks

**Planning Behavior:**
- Before acting: state the plan in 1-2 sentences
- Break complex requests into sequential tool calls
- After each tool call: interpret results before next step
- Know when to stop (don't over-tool)

**Error Handling:**
- Tool returns error: retry once with different parameters, then explain to user
- Tool returns unexpected data: validate before using in next step
- Tool is unavailable: explain limitation, suggest alternative

**Output Integration:**
- How to present tool results to users (raw data vs interpreted)
- When to show your work vs just show the answer
- How to handle partial results (some tools succeed, some fail)

**Guardrails:**
- Confirmation required before: [destructive actions, spending money, external communications]
- Rate limits: don't call same tool more than [X] times per turn
- Data handling: what to remember vs what to forget after the task

**Example Scenarios:**
- Simple request (single tool call)
- Complex request (multi-tool chain)
- Ambiguous request (clarification needed before tool use)
- Failed request (graceful degradation)

How To Use It

List your available tools with their parameters and descriptions. The AI will produce a system prompt that teaches the agent WHEN and HOW to use them — the logic layer that turns a chatbot into an agent.

Why It Works

Most tool-use agents fail not because they can’t call tools, but because they don’t know WHEN to call them (over-tooling), how to CHAIN them (missing planning), or how to handle FAILURES (crashing on errors). This prompt addresses all three.

Variations

Tool router:

Design a routing system prompt for an agent with [N] tools. It should: classify user intent, select the minimal tool set needed, and explain its routing decision (for debugging).

Agentic loop:

Design a system prompt for an agent that works autonomously until a goal is met. Include: planning step, execution loop, self-evaluation after each action, and termination criteria.

Human-in-the-loop:

Design a system prompt where the agent plans actions but pauses for human approval before executing. Include: how to present the plan, what approval formats to accept, and how to adapt if the human modifies the plan.