Starter prompts that earn their keep

The debug explainer

You are a senior engineer. Here is a failing snippet and its error.
Explain the root cause in one paragraph, then give the minimal fix.
Do not print the whole file back to me.

Role + context + output constraint. This pattern stops the "here is your code back" failure mode.

The structured writer

You are an editor. Rewrite my draft below.
Keep my tone, cut 30% of the words, and return:
1. the rewrite
2. three bullet points on what you changed and why
Draft: <your text>

Explicit output schema (> return exactly this) dramatically improves consistency across runs.

The researcher with a trap

Research this question and cite sources inline.
If you are unsure of a claim, mark it [UNSURE] and say why.
Never invent a citation — say "cannot verify" instead.
Question: <your question>

Anti-hallucination guardrails: force uncertainty to be visible and forbid fabricated citations.

Turn your chats into a prompt library

The premium prompts are the ones hiding inside your own history — the system prompts and custom instructions you refined over months. They live in your export as system-role messages. Run the analyzer to extract every custom instruction into one view, then search your history for the exact wording that worked. Reuse it, improve it, and store the good ones in your own notes.

Rule of thumb

Role → task → format → guardrail. If your prompt doesn't say what "good output" looks like, the model guesses. If it does, you get copy-paste quality on demand.