Prompt Engineering for Claude Code

★ 7.6 · vibe-coding

prompt-engineering helps Claude Code create, update, review, and diagnose LLM prompts — system prompts, skills, and agents. Every line of a prompt must close a real gap between natural model behavior and the required outcome; lines that no longer earn their place get pruned, so prompts stay balanced rather than accumulate over time. The skill branches on intent: creating (discover the goal and gap, draft the minimum), updating (locate each existing line's gap before patching), reviewing (check every line for necessity and edge-case robustness), and diagnosing failures (find the line driving the symptom before changing anything). Reference files — system-prompts, skills, knowledge-skills, agents, patterns, review, and metaprompting — load on demand only when their specific trigger fires. Useful for anyone writing or maintaining prompts for LLM-powered systems.