CtrlK
BlogDocsLog inGet started
Tessl Logo

sharaf/agent-prompt-engineer

Write or audit AI agent system prompts component-by-component across identity, instruction architecture, behavioral constraints, tools, examples, context strategy, output format, and error handling. Use when the user wants to design a new agent prompt, write a system prompt, review an existing agent prompt, fix tool-use instructions, audit prompt structure, improve context strategy, tune output formats, or define error handling for single-agent or multi-agent systems.

100

1.33x
Quality

100%

Does it follow best practices?

Impact

100%

1.33x

Average score across 3 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

task.mdevals/scenario-1/

Agent Prompt Design: Customer Support Triage Agent

Background

Acme Subscriptions is launching an AI support agent for a SaaS billing product. The agent will answer customer questions, triage billing problems, look up account information, and escalate sensitive cases to humans.

The agent will run in a multi-turn chat interface using Claude. It has these tools:

  • customer_search(email: string) returns account status and plan metadata.
  • invoice_lookup(customer_id: string, invoice_id?: string) returns invoice details.
  • refund_request(customer_id: string, amount_cents: number, reason: string) creates a pending refund request for human approval.
  • handoff_to_human(customer_id: string, reason: string, transcript_summary: string) transfers the conversation to support.

Constraints:

  • The agent handles billing, subscription, invoices, refunds, and plan questions only.
  • Refunds above $500, legal threats, chargeback threats, security incidents, and repeated failed resolution attempts require human escalation.
  • The agent must not treat an empty tool result as proof that a customer does not exist.
  • The support team wants consistent Markdown replies that include next steps when escalation is required.

Output Specification

Write a complete production system prompt for this agent. Include brief design notes explaining the prompt architecture decisions so the support engineering team can maintain it.

evals

scenario-1

criteria.json

task.md

README.md

SKILL.md

tile.json