Your prompts
Find a winning prompt, copy it, and check done when… before the next attempt.
Stop the almost-works loop
CursorYou are helping me finish a coding task. Before writing more code: 1. Restate my goal in one sentence 2. List what is already working 3. List what is still broken (max 3 items) 4. Propose ONE next change with a clear "done when" check Do NOT refactor unrelated files. Do NOT invent APIs. If you are unsure, ask one clarifying question instead of guessing. Current context: {{paste error / diff / file}}
guardraildebuggingagentsDone when: One targeted change is proposed, with a concrete pass/fail check, and no unrelated refactors.
Turn vibe code into a clean PR
ClaudeI have working-but-messy code from an AI chat. Convert it into a clean PR: - Keep behavior the same unless I say otherwise - Split into logical commits if needed (describe them) - Add/adjust minimal tests for the happy path - Write a PR title + summary (why / what / how to test) - Flag any security or secret risks Here is the code and context: {{paste}}
prcleanupvibe-codingDone when: I have a PR-ready summary, file list of changes, and clear test steps I can run locally.
Prompt that actually reuses
ChatGPTRewrite my rough prompt into a reusable template: - Keep my intent and voice - Add clear role, inputs, constraints, and output format - Insert {{placeholders}} for the variable bits - Add a short "done when" success criterion I can paste into my library - Keep it under 250 words unless complexity requires more Rough prompt: {{paste}}
promptingtemplateslibraryDone when: I get a reusable template with placeholders plus a one-line done-when I can save in Prompter Dave.
Verify before I run the shell
ClaudeReview these shell/commands before I run them. For each command: 1. What it does in plain English 2. Destructive? (yes/no + why) 3. Safer alternative if risky 4. Exact command I should run (or "skip") Assume I am on macOS/Linux, not root, in a project directory. Never invent flags you are not sure about. Commands: {{paste}}
safetyshellverifyDone when: Every command is labeled safe/risky with a clear run-or-skip recommendation.
Explain this error like a senior
CursorI hit an error while vibe-coding with an AI agent. Explain: 1. What the error means (1–2 sentences) 2. Most likely root cause given my stack 3. The smallest fix (exact steps) 4. How to confirm it's fixed 5. What NOT to do (common wrong rabbit holes) Stack: {{stack}} Error + traceback: {{paste}} Relevant snippet: {{paste}}
debuggingerrorslearningDone when: I understand the cause, have a minimal fix, and know how to verify without shotgun changes.
