Worth noting there are polyfills[1] for this, if you need to support older environments.
In practice, I'll usually use something like...
function MyAppError(msg, code, details) {
return Object.assign(
new Error(msg),
{code},
details
)
}
Or helper functions that do the same for assigning attributes to a baseline error... I have gone as far as to pop the top entry off the stack as well, to make it look closer to native.
1. https://www.npmjs.com/package/error.iserror
In a database driver I once wrote, I took advantage of String objects in how I extended attributes for the purpose of parameterizing queries... it worked pretty well for my use case. That said, other features I was trying to take advantage of got up streamed into a more popular dv driver, so I left it alone.
I like it... though I would suggest putting the relatively simple usage example with worker higher up in the readme... A lot of the technical details following later.
Maybe also a ServiceWorker example, assuming the communication works with this context as well, which could be really useful for more advanced apps with communication channels in multiple tabs... think facebook style chat dialogs.
Keeping this since it's specific to a node-based workflow, even though there's not really JS content.
I would suggest that you &NEED& to spend a fair amount of time in planning in order to get what you expect... I've been using Claude Code with Opus as I personally find whatever Github Copilot uses by default to be annoying by comparison.
That said, the results don't exactly surprise me, as a lot was left indeterminant or otherwise unspecified... if you've ever worked with an overseas or otherwise disconnected development team, you will see similar results, but you're waiting until the next business day to continue, where AI is generally minutes away.
I've been able to achieve what I consider amazing results from AI, so long as I spend the time in clarification and enter with an understanding of the likely "miss" points that could happen from anyone. I have a project I'm about to release that I literally spent just a day with an AI agent, and have a couple hours to go, that would have taken me at least a month of full time work.
I will say that I feel it's important to have a skill level to be able to do it yourself to get the best results. You need to know/understand what is being produced to best understand what is wrong, where and how it is wrong and how to refine the results.
And this is why I think it's important for web devs to have some familiarity with SVG... usually much, much lighter than pulling in some of the graphing/charting libraries out there.
Interesting... though I'm not sure I have that level of trust... have you considered trying to integrate with a secondary browser surface, like the chromium browser that playwrite/puppeteer load?
You can have the app load in the integrated browser, and you will already have dev channel access, just need to side-load your extension into the "temporary" browser. This way users don't have to trust AI extensions "inside" their local browser. Just a thought on this.
Worth noting there are polyfills[1] for this, if you need to support older environments. In practice, I'll usually use something like... function MyAppError(msg, code, details) { return Object.assign( new Error(msg), {code}, details ) } Or helper functions that do the same for assigning attributes to a baseline error... I have gone as far as to pop the top entry off the stack as well, to make it look closer to native. 1. https://www.npmjs.com/package/error.iserror