@IliyaOblakov good point on the 32 KiB limit, didn't know Codex cuts off there. gonna move the server vs client rule near the top. do you keep one AGENTS file or split per folder?
@olusegunka37813 An agent saying tests passed when they didn't is exactly why I'd want to see the real exit code. Does it also show which command the agent actually ran?
AI writes frontend code fast now, so do we still need tests? I think yes, but for confidence, not coverage.
For a Next.js blog, I'd keep it small: Vitest for the pure logic, Playwright for the list, detail and 404 pages. No snapshots.
What do you still check by hand?
@EthanLinAI Not having to re-explain the design system in every prompt is the real win. How well does it reuse existing components instead of making new ones?
@HimanshuRawal05 8 out of 101 makes sense. Half-done work usually comes with decisions nobody wrote down. Did the agents do better when the handoff had notes or a ticket?
@rewind02 Clicking through every flow is the part nobody talks about. Does it catch UI issues too, like a layout breaking on mobile, or mostly broken flows?
@eran_dror A fresh-context reviewer makes a lot of sense. The things you agreed on in chat but never wrote down are exactly what come back to bite you later.
@HimanshuRawal05 a test hitting a function the app never calls is such a sneaky one. save, reload, check it's still there should be the default for any form
before letting strangers talk to my AI chatbot, i ran a security checklist on it
tried 8 ways to break it. here's what i checked and what the audit caught ๐