Regarding what #2 said about "> So I was thinking I could just tell it to write tests, and my logic was it would scan my files...", yeah, it really doesn't work like that and you'll burn through your API budget so fast. I learned this the hard way last week. If you're on a tight deadline, I would suggest switching over to Cursor AI Code Editor Pro for this instead of the raw Claude CLI. Cursor runs on Anthropic models too but handles codebase indexing way better out of the box for 20 dollars a month, which is much cheaper than raw API billing when you're doing heavy context scanning. Just make sure to feed it a specific testing template for Vitest or Jest. If you need a good system prompt for React logins, let me know and I can paste mine here to help you get it sorted before Friday!

> So I was thinking I could just tell it to write tests, and my logic was it would scan my files and generate them automatically. But it just gives me these tiny one-liner tests that don't cover any edge cases. Yeah, unfortunately Claude Code out of the box is not as good as expected for this. I had issues with it failing to parse complex React contexts last week. I wasted hours trying to get it to mock an Auth0 login hook on my Dell XPS 15 9530 laptop and it just kept hallucinating the Jest syntax. If you want comprehensive coverage, you basically have to write a custom system prompt first with your exact mock data schemas. Honestly though, it might be faster to use Double.bot AI coding assistant VS Code extension. It has a dedicated test generation tool that actually handles mock injection correctly, unlike Claude which just guesses blindly.

Honestly, I totally get where you're coming from. I spent all last weekend fighting with the same issue while trying to finalize my portfolio site. It is super frustrating when these tools just output fluff. Unfortunately, relying on standard agents for complex React logic like auth flows has been hit or miss for me too. Before you burn your whole Friday, I found that using Vitest 1.6.0 combined with Testing Library React 16.0.0 gives way more stable results. I stopped asking the agent to just write tests from scratch. Instead, I open my project in VS Code 1.95.0 and have the agent build them incrementally. It’s way safer. I also started checking the official documentation at testing-library.com because it saves so much time compared to debugging AI-generated nonsense. Don't stress too much, you've got this!

Oh man, I ran into this exact same nightmare last month while rushing a release on my Apple MacBook Pro M3 Max 16-inch! Claude kept giving me those useless dummy tests that mock nothing useful. The absolute best trick here is using the context commands properly. You cant just ask it to write tests out of the blue. First, explicitly load your target files and your testing config using the add command, like `/add src/components/Login.tsx` and `/add src/hooks/useAuth.ts`. Once Claude actually has those in its context window, prompt it with a system-style instruction: 'Write Jest tests covering three specific scenarios: successful login, invalid credentials API error, and empty input validation.' Giving it a structured list of edge cases to target makes a massive difference! It went from writing one-liners to generating gorgeous, fully-mocked integration tests for me. It is absolutely amazing once you guide it!