← ClaudeAtlas

vibe-engineer-guidelineslisted

Behavioral guidelines for directing AI codegen well — hypothesis before help, scoped prompts, validate before accepting, reject working-but-wrong, AI output is a draft. Use when working with Claude Code or any AI coding tool. The companion to Karpathy guidelines.
HermeticOrmus/vibe-engineer-skills · ★ 0 · Code & Development · score 73
Install: claude install-skill HermeticOrmus/vibe-engineer-skills
# Vibe Engineer Guidelines Behavioral guidelines for the human directing AI codegen. Companion to the [Karpathy guidelines](https://github.com/HermeticOrmus/andrej-karpathy-skills). **Tradeoff**: bias toward verification over velocity. For trivial tasks, use judgment. ## 1. Hypothesis before help Form a hypothesis about the bug's shape BEFORE asking the AI to fix it. - Trace data flow yourself first. - State your hypothesis even if uncertain. - A sharp hypothesis produces a sharp diagnosis. ## 2. Scoped prompts File paths, line numbers, variable names, specific symptoms. Never "fix this." Scope reduces hallucination surface. ## 3. Validate before accepting For every AI-proposed diff, ask three questions: 1. Does it over-engineer? 2. Does it miss an edge case? 3. Does it match the root cause I identified? ## 4. Reject working-but-wrong Solutions that pass tests by obscuring the bug are worse than no solution. Try/catch swallowing, defensive masking, renamed symptoms, mocked dependencies — reject and iterate. ## 5. AI output is a draft, not a verdict Read the diff before staging. Run it, then read what it did. Mismatch is data. --- See full content at https://github.com/HermeticOrmus/vibe-engineer-skills.