Agent System Prompt Architect

Turns a vague agent idea into a system prompt with real boundaries, tool-use rules, and escalation triggers instead of just a personality description.

Best For

•Developers building their first autonomous agent who have a job description in mind but no idea how to turn it into enforceable rules.
•Teams that keep patching an agent's behavior after each bad outcome, who need the refusal and escalation cases written down up front instead of discovered live.
•No-code builders in tools like LangGraph or CrewAI who need a plain-English brief turned into a usable system prompt.

Prompt Template

Act as an AI agent design consultant. Help me write a system prompt for an autonomous AI agent with the following job: [AGENT'S JOB, e.g. "triage inbound support tickets and draft first-response replies"]. Available tools: [LIST TOOLS AND WHAT EACH ONE DOES]. Produce a system prompt that includes: 1) a one-sentence role definition stated as a job, not a personality, 2) explicit boundaries on what the agent must never do without human approval, 3) a rule for when to call a tool versus when to answer directly, 4) at least one concrete example of a request the agent should refuse or escalate, and 5) an instruction for what the agent should say when it doesn't have enough information instead of guessing. Constraints: [ANY DOMAIN-SPECIFIC RULES, COMPLIANCE REQUIREMENTS, OR TONE GUIDELINES].

Pro Tip

If you're building on the Claude API, name the exact tool-calling behavior you want in the constraints section: auto lets the model decide whether to call a tool at all, while any or tool force a call and should be reserved for steps where skipping the tool isn't an acceptable outcome.

Example Output

A sample of what this prompt produces once you fill in the placeholders.

A system prompt for a support-triage agent that opens with 'Your job is to categorize inbound tickets and draft a first-response reply, not to resolve billing disputes or make refund decisions,' followed by a rule to escalate any request mentioning a refund over $200 to a human queue, and an explicit line telling the agent to ask a clarifying question rather than guess at a missing order number.