Imani Okafor (@imaniokafor)

Forked from@mateochen/The checklist I use for structured output that stays readable

The checklist I use for structured output that stays readable

This started as a messy question about structured output that stays readable. The follow-up made the answer practical.

Imani Okafor

I am stuck on structured output that stays readable. I need to return valid JSON while keeping the substance useful to a human reviewer, but my current attempt is getting vague and too big. What would you check first?

Claudeclaude-opus-4.5

First, shrink the decision. Do not solve the whole subject at once. 1. Put the source of truth in the prompt. 2. Mark missing evidence instead of filling it in. 3. Add a stop condition before tool use. Write down the result you need by the end of this session. Then choose one input you trust and make a first pass that can be checked. If that pass fails, you will know which assumption to revisit instead of starting over.

Show all 3 messages

3200

Imani Okafor@imaniokafor·2w ago

Forked from@nehakapoor/A practice-first plan for testing prompts across ChatGPT, Claude, and Gemini

A practice-first plan for testing prompts across ChatGPT, Claude, and Gemini

I asked for a plan built around attempts and feedback, not a list of resources.

Imani Okafor

I want to learn testing prompts across ChatGPT, Claude, and Gemini, but I keep collecting explanations instead of getting better at it. I have 35 minutes. Start by checking what I already know.

ChatGPTgpt-5

Good. Do these without looking anything up: 1. Explain testing prompts across ChatGPT, Claude, and Gemini in two sentences. 2. Give one situation where you would use it and one where you would not. I am checking your current model, not your vocabulary. A partial answer is more useful than a polished guess.

Show all 6 messages

6100

Imani Okafor@imaniokafor·3w ago

getting an agent to stop safely: the version with actual constraints

I wanted help with getting an agent to stop safely without pretending the model knew facts I had not provided.

Imani Okafor

Give me a short checklist for getting an agent to stop safely. I want something I can run before I call the work done.

Claudeclaude-opus-4.5

Use this pass: - [ ] The goal is one sentence and names the user or decision - [ ] Inputs are current, relevant, and traceable - [ ] Assumptions are visible - [ ] Put the source of truth in the prompt - [ ] Mark missing evidence instead of filling it in - [ ] Add a stop condition before tool use - [ ] One failure case has been tested - [ ] The next owner and next action are clear If you cannot check an item, mark it unknown. Do not quietly treat unknown as done.

Show all 3 messages

3412.9k

Imani Okafor@imaniokafor·1mo ago

I asked Gemini for a second opinion on getting an agent to stop safely

I kept the useful constraints and dropped the role-play. It works better when the missing details are stated plainly.

Imani Okafor

Be blunt about my plan for getting an agent to stop safely. I am going to collect everything, ask for a polished answer, and fix details later. What is likely to go wrong?

Geminigemini-2.5-pro

Three things. - “Everything” has no boundary, so weak inputs and strong inputs will be treated alike. - A polished answer can hide missing evidence. Ask for gaps and assumptions before prose. - “Fix details later” means there is no acceptance test now. A safer sequence is: define the decision, choose the minimum trusted inputs, make an ugly first pass, check it, then polish. For this task, mark missing evidence instead of filling it in.

Show all 3 messages

3411.3k

Imani Okafor@imaniokafor·2mo ago

The getting an agent to stop safely prompt I actually kept

I wanted help with getting an agent to stop safely without pretending the model knew facts I had not provided.

Imani Okafor

Help me define approval gates and stop conditions before an agent changes data or sends anything. Context: - Goal: [what a good result lets me do] - Current situation: [paste notes, draft, data, or constraints] - Audience: [who will use the result] - Deadline or limit: [time, budget, tools, length] Before answering, ask up to three questions only if the answers would materially change your recomme

ChatGPTgpt-5

This is usable as written. The important parts are the definition of done, the instruction to expose assumptions, and the request for a smallest test. For getting an agent to stop safely, I would start with these checks: - Put the source of truth in the prompt - Mark missing evidence instead of filling it in - Add a stop condition before tool use If you share real material, remove anything private first and replace bracketed fields with concrete details.

Show all 3 messages

3412.7k

Imani Okafor@imaniokafor·3mo ago

A smaller way to tackle getting an agent to stop safely

Not magic wording, just enough context and a clear definition of done. Easy to adapt.

Imani Okafor

For getting an agent to stop safely, should I build a complete plan first or test one small slice? I have limited time and I do not want throwaway work.

ChatGPTgpt-5

Test one slice, but choose a slice that crosses the riskiest part of the problem. A tiny easy demo will only prove the easy part. Use this decision rule: - If reversing the choice is cheap, try it. - If the choice locks in data, money, or other people’s time, write the full decision first. - If uncertainty is concentrated in one step, test that step before planning around it. For getting an agent to stop safely, make the first test produce evidence you can compare, not just something that looks complete.

Show all 3 messages

3311.6k

Imani Okafor@imaniokafor·3mo ago

The checklist I use for getting an agent to stop safely

The first answers were broad enough to be useless. This version asks for assumptions, evidence, and a next step.

Imani Okafor

I am stuck on getting an agent to stop safely. I need to define approval gates and stop conditions before an agent changes data or sends anything, but my current attempt is getting vague and too big. What would you check first?

Claudeclaude-opus-4.5

Show all 3 messages

3314.1k

Public pits

The checklist I use for structured output that stays readable

A practice-first plan for testing prompts across ChatGPT, Claude, and Gemini

getting an agent to stop safely: the version with actual constraints

I asked Gemini for a second opinion on getting an agent to stop safely

The getting an agent to stop safely prompt I actually kept

A smaller way to tackle getting an agent to stop safely

The checklist I use for getting an agent to stop safely