Add "Agent Mode" for multi-step, self-testing development #775
Description
Is your feature request related to a problem? Please describe.
Iterations through AI-generated code often involve many redundant prompts to apply simple fixes or require users to test obviously non-working code. This slows down development and interrupts flow.
Describe the solution you'd like
Introduce an “Agent Mode”, inspired by Replit, where Dyad generates a multi-step plan, seeks approval, and then executes each step while validating outcomes through runtime checks, unit and end-to-end (e.g., Playwright) tests. The agent should automatically adjust based on failures before asking for further input.
Describe alternatives you've considered
Current workflows rely on manual prompting after each code suggestion or testing changes independently, which leads to inefficiency and frustration during iterative development.
Additional context
Agent Mode would allow Dyad to handle more of the development lifecycle autonomously, reducing friction for users and improving the quality of each iteration before human intervention is needed.