Agent Starter Kit

You are a senior AI systems engineer. I'm designing an agent that does: [describe task].

Walk me through every failure mode before I write any code:
1. Loops — where could the agent get stuck repeating itself?
2. Irreversible actions — what could it do that can't be undone?
3. Lost state — what happens if it crashes halfway through?
4. Silent errors — where could it fail without telling me?

For each failure mode, give me the specific guardrail I need to build.

I'm building an agent that needs memory. Give me a decision framework for:

1. What belongs in short-term context (the prompt window)?
2. What belongs in long-term storage (a database)?
3. What should be retrieved on-demand via tools?

Then apply that framework to my agent: [describe what your agent does and what it needs to remember].

Flag anything that will hit token limits fast and what to do about it.

I need to design the tools for my agent. Here's what it should be able to do: [list capabilities].

For each capability, give me:
1. The exact tool signature (name, parameters, return type)
2. What can go wrong when this tool is called
3. Whether the action is reversible or not
4. The specific guardrail or validation I need before executing

Then tell me which tools I should build last — the ones most likely to cause real damage.

Before I build this agent, help me design its eval suite. The agent does: [describe your agent].

Give me 10 test cases covering:
- The happy path (3 cases)
- Edge cases the agent will definitely encounter (4 cases)
- Adversarial inputs designed to break it (3 cases)

For each test case: the exact input, the expected behavior, and how I can verify it automatically without a human in the loop.

My agent is ready to ship. Run me through a production readiness checklist.

For each category below, ask me the specific question I need to answer before going live — then tell me what the right answer looks like:

1. Observability — can I see what the agent is doing in real time?
2. Rate limits — am I protected from runaway loops burning my budget?
3. Fallbacks — what happens when a tool call fails or times out?
4. Cost controls — is there a hard cap on spend per session?
5. Rollback — can I instantly disable this agent if something goes wrong?

An agent I deployed is doing something unexpected: [describe the behavior you're seeing].

Help me respond in order:

1. Root cause — what are the 3 most likely explanations for this behavior?
2. Diagnostics — what specific information do I need to gather right now?
3. Containment — how do I stop the bleeding without losing important state?
4. Fix — given the most likely root cause, what's the minimal change that resolves it?
5. Post-mortem — what do I add to my monitoring so this surfaces faster next time?