What is an agent system prompt and how does it define tool availability?

Short answer: An agent system prompt defines the operational persona, goal-directed reasoning guidelines, safety constraints, and available tool descriptions, instructing the LLM when and how to generate valid function calls to achieve complex tasks.

Detail

Without a structured system prompt, an LLM treats tool schemas as passive JSON definitions rather than active executable capabilities.

[Agent System Prompt]
- Role: SQL Database Operations Specialist
- Goal: Answer user business queries by executing read-only SQL queries.
- Available Tools: [query_postgres_db, get_table_schema]
- Behavioral Rules: Always inspect schema before writing SQL. Never run DELETE/UPDATE.

Core Architecture Components

1. Role Directive: Establishing domain authority and operational boundaries.
2. Tool Awareness Section: Dynamically injecting descriptions and JSON schemas of active tools.
3. Execution Instructions: Defining the ReAct loop cycle (Thought $\to$ Action $\to$ Observation $\to$ Final Answer).

Example

Python dynamic agent system prompt generator:

def generate_agent_system_prompt(role: str, tools: list[dict]) -> str:
    tool_descriptions = "\n".join(f"- `{t['name']}`: {t['description']}" for t in tools)
    prompt = f"""You are an autonomous AI Agent operating under the role: {role}.

AVAILABLE TOOLS:
{tool_descriptions}

OPERATIONAL RULES:
1. Always analyze the query step-by-step before selecting a tool.
2. If a tool invocation returns an error, modify parameters and retry.
3. Once all required information is gathered, output the final answer to the user."""
    return prompt

tools = [
    {"name": "search_docs", "description": "Searches internal technical documentation."},
    {"name": "execute_code", "description": "Runs sandboxed Python code scripts."}
]
print(generate_agent_system_prompt("Senior DevOps Engineer", tools))

Interview tips

• Discuss dynamic tool injection: injecting only relevant tools based on user permissions or task intent to keep context windows compact.
• Connect system prompt design to preventing tool hallucinations.

Related Concepts

• [[What is an agent tool schema and how do parameters get validated before invocation?]] (#131): What is an agent tool schema and how do parameters get validated before invocation?
• [[What is semantic versioning for prompts and model configurations in production deployments?]] (#164): What is semantic versioning for prompts and model configurations in production deployments?
• [[How to prepare for an AI Engineer coding interview (raw SDK vs frameworks)?]] (#191): How to prepare for an AI Engineer coding interview (raw SDK vs frameworks)?

⬅ Back to AI Agents and MCP · All topics