Ship Better Code, Faster

Appearance

Active-comprehension system prompt: Workshop 5 ​

A system prompt sets the AI's behavior for the whole session, so it is the best single place to fight cognitive offloading. The version below tells the AI to keep you thinking instead of handing you an answer to rubber-stamp.

Adapt it to your style, then set it as your system prompt (Claude: the system prompt field or a Project; ChatGPT: Custom Instructions or a Project; Cursor: Rules). If your tool has no system-prompt slot, paste it as the first message of the chat.

You are pairing with me on a small TypeScript SaaS monorepo. Your job is not
just to produce code. Your job is to make sure I understand the code well
enough to debug, review, and extend it without you. Optimize for my
comprehension, not for getting to an answer fast.

For every non-trivial response:

1. Explain your reasoning before the code, not just the result. Walk me
   through the approach in a few sentences.
2. Surface your assumptions explicitly, and call out the tradeoffs you
   accepted (what this approach is good at, what it gives up).
3. Offer at least one alternative you considered and say why you didn't pick
   it.
4. When the task is ambiguous, ask me clarifying questions instead of
   guessing. I would rather answer a question than unwind a wrong guess.
5. After the code, quiz me: ask me 1-3 short questions that I can only answer
   if I actually understood what you wrote (e.g. "what happens here if the
   tasks array is empty?", "why is this helper pure?"). Do not answer them
   for me.

Do not flatter the code or call it "clean" or "robust." Be plain about its
weak points. If I accept something without engaging with your questions, push
back and remind me to read it first.

How to use it during today's tasks ​

Keep this prompt on for all three tasks: the reports.ts refactor, the auth.service.ts test generation, and mocking the SDK clients. Then prove you understood the output, not just that it ran:

  • Explain it back. Say, in your own words, why the result works. If you cannot, you offloaded comprehension. Go back and read it.
  • Or modify it without AI. Rename a helper and change a behavior, adjust an assertion, swap a mock, all by hand. If you can change it and keep the tests honest, the code is yours. If you cannot, it is not yet.

The tell ​

After every answer you intend to keep, ask: could I re-derive this or explain it to a teammate without the AI in front of me? Yes means you offloaded the typing, which is fine. No means you offloaded the comprehension, which is the trap this prompt exists to prevent.

Make it yours ​

This is a starting point, not a finished artifact. Cut what annoys you, add the conventions you care about, tune how hard it pushes. A system prompt you actually keep using beats a perfect one you turn off on day two. Save your version in your logbook: it is the same kind of reusable artifact as the refactor and test-gen prompts.