Many online resources treat VSCode like an editor with Claude bolted on. I use it the other way around. Claude does most of the writing. VSCode is where I catch it when it drifts. I'm not a developer. My Master's degree is in curriculum design. I run two active projects out of VSCode, a Next.js site for my digital products brand and a NestJS accounts receivable backend for a solar company. Both live and die by whether Claude follows the rules I set. VSCode is how I make those rules visible. A few things I do that I don't see talked about much. Pin the rulebook, not the code. I keep CLAUDE.md, the current task file, and my master data CSV open in split panes before I start a session. When Claude drifts I can see exactly which rule got ignored. If it's already open in the sidebar errors become obvious. Use the diff view as a pause button. After Claude makes changes I never let a commit happen until I've scrolled through the git diff in VSCode. This has caught hiccups and allowed for guardrails to prevent future issues. Search across files to audit for drift. My brand has hard rules on things like em dashes and specific capitalization for divine pronouns. Audits catch AI drift the same way spellcheck catches typos. Verification scripts as VSCode tasks. I have small Python scripts that count rows and check character limits. Running them from the terminal was fine, but making them one-click tasks changed how often I actually use them. Now every session ends with a task run. If it doesn't pass, the work isn't done. Have you noticed your AI doesn't always tell the truth? Make it verify its work. Multi-root workspaces to keep projects separate. I'm sure you all already to this, but it's worth saying. My AR backend and my devotional brand are very different contexts. Keeping them in separate workspaces means their CLAUDE.md files don't accidentally cross wires. None of this is fancy engineering. Goodness knows, I don't have a tech background. It's mostly refusing to trust the AI's report on itself and using VSCode where I can see the work done.

I hope some members here have some UI design experience and have run into this issue before. So I use Claude Design and Open Design to create layouts for the web apps I build. Both tools generally design the interfaces with React Components. All cool, I use these as prototypes. The problem is I do not use React in my projects so I need to convert these designs to my stack. And this is were I run into a problem. No matter what I do during the conversion the designs are approximated and are not the same design by a long shot. I am stuck. If I can't figure out the prompts to make this work I cannot setup a ICM system to make this work reliably. My experience is literally the reference design is just a suggestion and not a source of truth no matter what prompts I use. This is the same for Codex and Claude Code. How do I fix this process? Anyone has encountered this before?