Replace hardcoded Playwright use

[LotemAm]

This tool is great and I've used it to develop a Chrome extension, and now I'm using it to develop a desktop application. For both of these use cases the Playwright specification posed a problem, even when developing the Chrome extension, the agents couldn't figure out how to use the Playwright config that loads the extension.

I think it would be better if the Initializer Agent decided what testing framework a project should use, and configure the project accordingly.

[Bortus-AI]

I have the same issue with Playwright and autoforge

…

On Thu, Feb 19, 2026, 1:58 PM Lotem ***@***.***> wrote: *LotemAm* created an issue (AutoForgeAI/autoforge#209) <#209> This tool is great and I've used it to develop a Chrome extension, and now I'm using it to develop a desktop application. For both of these use cases the Playwright specification posed a problem, even when developing the Chrome extension, the agents couldn't figure out how to use the Playwright config that loads the extension. I think it would be better if the Initializer Agent decided what testing framework a project should use, and configure the project accordingly. — Reply to this email directly, view it on GitHub <#209>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AX42UPAGCSRV33QCXTK4MA34MYIWFAVCNFSM6AAAAACVXRD2O2VHI2DSMVQWIX3LMV43ASLTON2WKOZTHE3DIOBRGAZTAMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[m13v]

this is a real gap. Playwright is great for browser testing but it simply can't drive a Chrome extension's popup UI or interact with desktop app windows. for desktop applications, the OS accessibility APIs are the right tool: Windows UI Automation for Windows apps, AXUIElement for macOS. these give you the same kind of element tree that Playwright provides for web pages (elements with roles, labels, bounding rects), but for native app UIs.

for Chrome extensions specifically, the extension popup renders in a separate browser context that Playwright can't easily attach to. the workaround is either loading the extension in a test page (which changes behavior) or using the accessibility tree to find and interact with the popup window directly.

having the Initializer Agent pick the testing framework based on the project type is the right idea. the heuristic could be: if the project has a manifest.json (extension) or electron-builder config, use accessibility-based testing. if it's a standard web app, use Playwright. if it's a hybrid, use both.

[m13v]

our desktop automation framework that provides the accessibility tree-based testing I described, covering both Windows (UI Automation) and macOS (AXUIElement) with a unified element model: https://github.com/mediar-ai/terminator/blob/main/crates/terminator/src/element.rs

installable as an MCP server for AI agents via npx: https://github.com/mediar-ai/terminator/blob/main/crates/terminator-mcp-agent/src/server.rs