I built a similar system for an identity protection service that automated removing PII from directory websites like whitepages. Which was less ethical, stealth browser automation or monetized privacy invasion?
other people in the thread say that posix_spawn is more or less implemented as a fork+exec wrapper though? it sounds like the idea is more like if there were a separate deferred_fork that made an intermediate "process factory" that let you set up a process without actually creating a new one until the exec. obviously the if() construct would have to be replaced with an in-process handle that mimics calls to the posix api.
We use claude -p and a max subscription for our devtools, this seems like a sufficient workaround to avoid usage based pricing. I had already tried letting max users sign into a web frontend with oauth, but that was obviously not an anthropic sanctioned use case.
With a vibecoded multiplayer web IDE for our flutter app, I have two nontechnical team members prototype app features from a browser all day. The outcome is a git branch with functional dart code, and a strictly more powerful design prompt than figma make. Claude built its own claude -p harness that communicates with the browser via websocket flawlessly. To use our max subscription after [1] takes effect, we are considering adapting that setup to an approach like [2] instead.
yep, it generates 700 tools on my mac. I consider discovery context a separate problem, there is no more reasonable way to ask an MCP to limit the number of tools it exposes IMO, than there is to ask a linux distribution to paginate /bin. for the context problem, you would want the harness to use vector search on the list of tools to find the ones relevant to the given prompt instead. not sure if any do this currently.