Acceptance criteria, a running app, and controlled test data.
Checks for your critical journey, shown to catch the persistence defect, with what remains untested written down.
Prove the shared persistence defect is detected, then verify the repaired user journey and record remaining unrun checks.
Planned: coming to the library.
A green build does not show that saving works. This skill guides your coding agent to connect your acceptance criteria to unit, browser and manual evidence, so you know what was actually checked before you share the app.
Checks are chosen by the failure they must detect: unit checks for a validation rule, integration checks for a write followed by a fresh read, browser checks for the real journey through the interface, and manual checks for wording, keyboard access and layout. The agent covers persisted create and read, invalid input and failed writes, plus edit and delete where they exist, with isolated data and accessible selectors.
The regression check has to fail against the broken behavior and pass after the repair, so you know it can catch the defect. Manual checks and checks blocked by a missing service are reported separately, not silently turned into skips. It reuses your project's existing test runner.
You need a coding agent with access to your project files. Provider accounts, model weights, GPU compute and hosting are not included.
No reviews yet.
Sign in to rate this Vibe.
Make the failure happen on demand, fix the one cause behind it, and prove the same steps now pass, without guessing in circles.
Model tasks, apply a development migration, and verify fresh reads after restart.
Add accounts and prove that each person can access only their own tasks.
Prepare Vercel or Railway, then check the running journey, persistence, and recovery.