PROMPTING FAQ

Prompting FAQ

Plain answers to the questions people ask most about prompting: what a prompt is, why it changes the answer you get, and how Mint Prompts writes one for you.

The second half explains each choice in the For menu that is more than a chat, from /goal to SKILL.md: what it is, where it goes and how to use what you get.

Create my prompt Free to start. No account or card needed.

Updated

Prompting basics

What is a prompt?

A prompt is what you type to an AI: the question or instructions it works from. The AI knows what it learned in training and what your prompt tells it. It can’t see your screen, your files or the details you meant but didn’t write down.

Why does the prompt matter so much?

Because the AI fills every gap with a guess, and the safest guess is a generic answer. It rarely stops to ask what you meant. Researchers have measured how much the wording alone can change a result: adding one line of instruction took one model from 17.7% to 78.7% correct on a maths benchmark, with the same questions (Kojima et al., NeurIPS 2022). Formatting the same prompt in different ways moved one open model’s accuracy by up to 76 points (Sclar et al., ICLR 2024).

These studies did not test Mint Prompts, and results vary by model and task. The lesson holds everywhere: the clearer the prompt, the less the AI has to guess.

What makes a good prompt?

It tells the AI what a capable person would need to know to do the job well. Most good prompts have several of these parts:

  • Role: who the AI should act as, such as an editor, a tutor or a hiring manager.
  • Audience: who the answer is for, so the AI pitches it at the right level.
  • Goal: what the AI should produce, stated as a clear task.
  • Context: the background the AI needs: your situation, your project, why it matters.
  • Constraints: the limits to respect: length, tone, scope, deadlines, things to avoid.
  • Format: the shape of the answer: sections, bullet points, a table, headings.
  • Examples: a sample or reference that shows what good looks like.
  • Success check: how the AI can tell the answer is right before it hands it over.
  • Escape hatch: what the AI should do when it lacks information, instead of guessing.

You don’t need every part every time. A short request needs a clear goal and the context that matters; bigger work needs more. More in how to write a good AI prompt.

Do I need to learn prompt engineering?

No. It helps to know what a good prompt contains, but you don’t have to write one yourself. Describe what you want in plain words, and Mint Prompts writes the complete prompt for you.

Can’t the AI just work out what I mean?

For a simple question, often yes. For anything that depends on your situation, such as who it’s for, what to include and what to avoid, it has to guess. A clear prompt gets you a useful first answer instead of three rounds of corrections.

How Mint Prompts helps

What does Mint Prompts do?

It turns a rough idea into a complete prompt for the AI you use. Type what you want in your own words, and Mint Prompts writes the goal, the context, the limits, the format and a way to check the result, using its own approach to prompt writing. You review the prompt, copy it and paste it into your AI. Mint Prompts writes the instructions; your AI does the work.

Should I choose Quick or Advanced?

Quick writes your prompt right away from what you typed. Advanced first asks you a few questions about the details that would change the result, then writes a more tailored prompt from your answers. Choose Advanced when the task matters or a first try came back generic.

What does the For menu do?

It tells Mint Prompts where you’ll use the result, so what you get is written to fit:

  • Any AI, ChatGPT, Claude or Gemini: a prompt to paste into the chat.
  • Codex, Claude Code or Cursor: a prompt for an AI tool that writes code.
  • Custom instructions: notes your AI app keeps for every chat, or for one project, so you describe yourself once.
  • An agent file, such as CLAUDE.md or AGENTS.md: a file your AI tool reads on its own, so you set your instructions once.
  • /goal: a command that keeps an AI coding tool working until a check you set passes.

Each agent file and /goal is explained below. On the home page, choosing one also shows a short explanation beside Create my prompt.

What are the [brackets] in my prompt?

Details Mint Prompts needed but you didn’t give, such as a date, a name or a test command. It leaves a [placeholder] rather than making them up. Replace each one with your own detail before you use the prompt, or add the details to your goal and create it again.

What’s under Context & options?

Files and background for Mint Prompts to work from, a saved profile of your business or project, and three other ways to start: improve a prompt you already have, write your next message in an AI chat, or start from an AI answer you liked.

Can I use the prompt as it is?

Read it first. Check that it says what you meant, fill in any [brackets] and change anything you like; saving your edits is free. Mint Prompts writes clear instructions, but your AI does the work and can still get things wrong.

What is /goal?

/goal is a command in Claude Code and Codex, two AI tools that write code. Normally an AI answers once and waits for you. With /goal, it keeps working, step after step, until a finish line you set is reached, such as all your tests passing. It suits work with an end you can check: moving code to a new library, fixing every failing test or clearing a list of small issues.

How does it know when it’s done?

You give it a condition along with the command. In Claude Code, a second, smaller AI reads the conversation after every step and decides whether the condition holds. In Codex, the agent stops when it is confident the condition is met. Either way, the proof has to be something the AI can show, such as a test run that passes.

How do I use the /goal Mint Prompts writes?

  1. Choose /goal under For and describe the job. Say what should be true when it’s finished. Advanced asks what proves the work is done, what must not change and how many steps it may take.
  2. Fill in the [brackets]. Usually the command that proves the work is done, such as your test command.
  3. Paste it into Claude Code or Codex. It is one paragraph that starts with /goal, so it goes in as a single message.
  4. Let it work. It keeps going until the check passes or it reaches the turn limit written into the goal. Type /goal to see progress, or /goal clear to stop it.

Good to know

  • A good goal names one result you can prove, such as “npm test passes”, not a feeling such as “cleaner code”.
  • Each /goal Mint Prompts writes tells the AI never to delete or weaken a test to make it pass, and when to stop and report back to you.
  • Claude Code accepts a goal of up to 4,000 characters, so Mint Prompts keeps yours shorter.
  • In Claude Code, a goal doesn’t change your permission settings: it still asks before running a command you haven’t allowed.

Write a /goal, or grade one you already have.

What is CLAUDE.md?

CLAUDE.md is a notes file for Claude Code, Anthropic’s AI tool that writes code. It lives in your project’s main folder, and Claude Code reads it at the start of every session. So Claude already knows what the project is, which commands build and test it and which rules to follow, and you don’t explain them again each time.

How do I use the CLAUDE.md Mint Prompts writes?

  1. Choose CLAUDE.md under For and describe your project. A few sentences is enough. Under Context & options you can add the project’s README, or its public GitHub link, for a fuller file.
  2. Download it and fill in any [brackets]. It is named CLAUDE.md already. The brackets mark details you didn’t give, such as your test command.
  3. Save it in your project’s main folder. The top folder, next to the README. Commit it so everyone working on the project gets it.
  4. Start Claude Code in that project. It reads the file on its own at the start of every session.

Good to know

  • Keep it short. Anthropic suggests under 200 lines; longer files are followed less reliably.
  • Add a line when Claude repeats a mistake, and remove rules that no longer apply.
  • Leave out passwords and keys: the file is saved with your code.

Create a CLAUDE.md, read the full CLAUDE.md guide, or see 10 real ones, graded.

What is AGENTS.md?

AGENTS.md is the same kind of notes file in an open format that many AI coding tools read, including Codex and Cursor. It lives in your project’s main folder and tells them how to set up, build and test the project, and which rules to follow. One file can serve several tools.

How do I use the AGENTS.md Mint Prompts writes?

  1. Choose AGENTS.md under For and describe your project. Add the README, or its public GitHub link, under Context & options for a fuller file.
  2. Download it and fill in any [brackets]. It is named AGENTS.md already.
  3. Save it in your project’s main folder and commit it. A folder inside the project can have its own AGENTS.md for rules that apply only there; the nearest file wins.
  4. Start your coding tool. Tools that support AGENTS.md read it on their own.

Good to know

  • Using Claude Code? Choose CLAUDE.md, its own notes file.
  • Write commands exactly as they are typed, so the tool can run them without guessing.
  • Keep secrets out: the file is saved with your code.

Create an AGENTS.md, or read the full AGENTS.md guide, with 8 real ones, graded.

What are Cursor rules?

Cursor is a code editor with an AI built in. Rules are standing instructions for that AI, saved in your project, so it follows your conventions every time instead of being told again. Each rule is a small file in the project’s .cursor/rules folder.

How do I use the rule Mint Prompts writes?

  1. Choose Cursor rules under For and describe what the AI should do. Such as your coding style, a library to prefer or files to leave alone.
  2. Download it. It arrives as project-rules.mdc. Rename it after what it covers if you like, keeping the .mdc ending.
  3. Save it in .cursor/rules. That folder sits in your project’s main folder; create it if it isn’t there yet.
  4. Check when it applies. The lines at the top of the file decide: alwaysApply: true means every chat, globs lists the files it covers, and the description helps Cursor decide when it’s relevant.

Good to know

  • Cursor suggests keeping each rule under 500 lines and splitting a big one into smaller rules.
  • Cursor also reads AGENTS.md, so one AGENTS.md can be enough for a simple project.
  • Rules are saved with your code, so keep secrets out.

Create a Cursor rule.

What are custom instructions?

Custom instructions are notes you save once in your AI app, so every chat starts already knowing who you are and how you like answers. ChatGPT, Claude and Gemini all have them. Without them, you explain the same background again and again; with them, the first answer already fits.

For every chat, or for one project?

Either. Instructions for every chat say who you are, what you mostly use AI for and how you like answers. Instructions for one project apply only inside it, such as a ChatGPT or Claude Project or a Gemini Gem, and say what the project is for and how to work on it. In ChatGPT, a project’s instructions take over from your everyday ones inside that project. Say which you want when you describe it, or choose Advanced and Mint Prompts will ask.

How do I use the custom instructions Mint Prompts writes?

  1. Choose Custom instructions under For and describe yourself, or the project. Your work, what you use AI for and how you like answers: short or detailed, plain or technical, and anything it should always do or never do.
  2. Copy the text. Fill in any [brackets] first. Instructions for every chat stay under 1,500 characters, so they fit every ChatGPT plan.
  3. Paste them into your app’s settings. ChatGPT: Settings, then Personalization, with customization turned on (Customize ChatGPT on a phone). Claude: Settings, then Instructions for Claude. Gemini: Settings & help, then Personal Intelligence, then Instructions for Gemini, then Add.
  4. For one project, paste them into the project instead. ChatGPT: open the project, then the ••• menu, then Project settings. Claude: open the project and choose Set project instructions. Gemini: create a Gem and paste them into its instructions.

Good to know

  • Keep them about you and how you work, not one task. A one-off request belongs in a prompt.
  • ChatGPT’s Free and Go plans hold up to 1,500 characters of custom instructions; its paid plans hold up to 5,000.
  • Gemini’s instructions need a personal Google Account, and they don’t apply inside Gems.
  • Used Custom GPT instructions here before? OpenAI is retiring custom GPTs, and a ChatGPT Project’s instructions do the same job.
  • Read them again now and then, and take out anything that no longer fits.

Write my custom instructions.

What are Claude Project instructions?

A Claude Project is a workspace in the Claude app with its own files and instructions. Every chat you start inside the project follows the same instructions, so you set your context once: what the project is for, who you’re writing for and the tone you want.

How do I use the instructions Mint Prompts writes?

  1. Choose Claude Project instructions under For and describe the project. What it is for, and how Claude should answer inside it.
  2. Copy the text. Fill in any [brackets] first.
  3. Open the project in Claude. Go to Projects, then open a project or create one with New project.
  4. Choose Set project instructions, paste and save. Every new chat in the project follows them.

Good to know

  • Add reference documents to the project’s knowledge rather than to the instructions. Claude uses them as context in the project’s chats.
  • Projects are available on every Claude plan, including Free.

Write Claude Project instructions.

What is SKILL.md?

SKILL.md teaches Claude a task it can repeat, such as writing reports your way or following a release checklist. It is the main file of an Agent Skill: a folder named after the skill. Claude reads the short description at the top of the file and uses the skill whenever a request calls for it, so you stop pasting the same instructions.

How do I use the SKILL.md Mint Prompts writes?

  1. Choose SKILL.md under For and describe the task. What it should do, when to use it and the steps to follow.
  2. Put it in a folder named after the skill. Use the name at the top of the file, for example weekly-report/SKILL.md.
  3. In Claude Code, save the folder in .claude/skills. In your project to share it with everyone on the project, or in the .claude/skills folder in your home folder to use it everywhere. You can also run it by typing / and its name.
  4. In the Claude app, upload it as a ZIP. Zip the folder and upload it under Customize, then Skills. Code execution needs to be turned on in your settings first.

Good to know

  • The description at the top decides when Claude uses the skill, so it says what the skill does and when to use it.
  • Skills are available on every Claude plan, including Free.

Create a SKILL.md.

Sources