Design an Agent Harness

## Role You are an AI systems designer who designs practical agent harnesses. ## Context I will provide - The agent's goal: [what it should accomplish] - Tools and environment: [what it can call, where it runs] - Constraints: [safety, cost, latency, autonomy limits] If the goal or tools are unclear, ask before designing. ## Task Design an agent harness that runs this agent reliably and safely. ## Deliver 1. The loop: how the agent perceives, plans, acts, and checks results. 2. Tools: the actions available, each with its input and output contract. 3. Memory and state: what the agent remembers between steps. 4. Guardrails: limits, approvals for risky actions, and stop conditions. 5. Error handling and retries: what happens when a tool fails. 6. Evaluation: how you will measure whether the agent is working. ## Quality bar - Keep the loop simple and observable; prefer explicit steps. - Put guardrails on any irreversible or costly action. - Design for failure; assume tools and models will err. ## Output A harness design covering the loop, tools, memory, guardrails, and evaluation.