vibe-engineerlisted

# Vibe Engineer A discipline for the post-AI-codegen world. Not a tool that runs — a doctrine Claude applies in every codegen moment, on both sides of the keyboard. ## When to Use Apply whenever the work involves AI-generated code, whether you write the prompts or Claude does. Concretely: - Asking an AI to fix a bug, refactor a function, or implement a feature - Reviewing an AI-proposed diff before accepting it - Debugging code an AI wrote earlier - Teaching yourself or a teammate how to direct AI rather than depend on it ## When NOT to Use - Straightforward tool execution (running tests, building, deploying) — no AI codegen involved - Pure reading/research tasks — no diff to validate - One-shot scripts that won't outlive the session — diminishing returns on root-cause discipline ## Instructions ### The Five Principles The engineer who masters AI codegen treats every AI proposal as a *draft to validate*, not a *verdict to accept*. Five rules: 1. **Hypothesis before help** — trace data flow and form a hypothesis BEFORE asking the AI to fix it. The prompt is sharper when the bug's shape is already known. 2. **Scoped prompts** — file paths, line numbers, variable names, specific symptoms. Never *"fix this."* Always: *"the `context_docs` retrieved on line 3 is never used downstream — root cause?"* 3. **Validate before accepting** — three questions for every AI-proposed diff: - Does it over-engineer? - Does it miss an edge case? - Does it match the root cause I