1. Tell it to keep going until done GPT-5.1 sometimes quits early. Add: "You're biased for action. If I ask 'should we do X?' and the answer is yes, just do it. Don't stop halfway - finish the whole task before asking me anything."

2. Give status updates every 6-8 actions Make it report progress: "Send me a 1-2 sentence update every 6 steps or 8 tool calls. Tell me what you found or did, not just what's next."

3. Respond fast, think later For better user experience: "Always post a quick message explaining what you're doing FIRST, before you start your deep thinking."

4. Control how much it talks Be specific: "Keep responses to 2-5 sentences for small tasks. Only write more for complex multi-step projects." You can also use the verbosity parameter.

5. Define its personality upfront Tell it exactly how to act: "Be direct and efficient. Match my energy—if I'm brief, you're brief. Skip 'got it' and 'thank you' unless I'm being really warm."

6. Explain WHAT tools do in the definition, WHEN to use them in the prompt In tool description: "This creates reservations." In prompt: "MUST call this tool when user asks to book. Never guess - ask for missing info."

7. Make it use multiple tools at once Add: "Parallelize tool calls whenever possible. Run multiple searches or file reads at the same time to go faster."

8. For coding: Use a plan tracker Create 2-5 big milestones (not tiny steps like "open file"). Keep ONE item as "in progress" at a time. Update it every ~8 actions.

9. Use "none" mode for simple tasks When you don't need deep reasoning, use none mode BUT add: "Plan carefully before calling any function. Verify your outputs. Keep going until fully done."

10. Debug bad prompts by asking GPT-5.1 Feed it your prompt + examples of failures. Ask: "What contradictions do you see?" Then ask it to fix those specific issues with small edits—don't rewrite everything.