Stop Telling Claude Code What To Do

Description

📝 Prompts available on GitHub: https://github.com/glittercowboy/taches-cc-prompts.git 🙋🏼‍♂️ Join live Q&A calls with me once a week for just $57/m (7-day free trial): https://claude-coders.circle.so/checkout/claude-coders A walkthrough of how I use "meta-prompting" with Claude Code—where instead of directly telling Claude what to do, I first ask it to create an optimal prompt for the task. I demonstrate a custom /create-prompt command that acts as a prompt engineer. It analyzes your request, asks clarifying questions if needed, structures the prompt with XML tags, and saves it to a prompts folder. Then a /run-prompt command executes these as fresh sub-agents with clean context windows. The comparison shows building a generative music CLI tool with and without this approach. The meta-prompted version produces better code organization (modular vs monolithic), asks useful clarifying questions upfront, and includes built-in verification steps. Worth noting: In retrospect, the single-shot prompt would've been fairer if I'd given it the same clarifying context I provided to the meta-prompt version. But the core advantage still holds—forcing Claude to think about how it should be asked to do something produces more thoughtful, structured results for complex tasks. This is 100% overkill for simple changes but I've found it be essential for any substantial work I do like big refactoring tasks, database migrations, or multi-file features that require incredibly specific instructions. 0:00 Intro 00:56 The 'Create Prompt' Prompt 13:36 The 'Run Prompt' Prompt 14:23 Using Meta-Prompting in Practice 29:09 Outro