Prompt Chaining for Agent Construction

Step 1: "Should this be an agent?" [Section 1]
Step 2: "Specify the agent." [Section 1]
Step 3: "Design the tool surface from the tasks." [Section 3]
Step 4: "Write the schemas." [Section 3]
Step 5: "Review the schemas hostilely." [Section 3]
Step 6: "Write the system prompt." [Section 2]
Step 7: "Enumerate failure modes." [Section 1]
Step 8: "Which failure modes need code, not prompt?" [Section 2]
Step 9: "Design guardrails for the rest." [Section 8]
Step 10: "Build the eval set." [Section 7]