The vulnerabilities kinda rely on having to make quite a number of requests to exploit + can crash the process as I understood, so if many people try to force their way in at once 2h doesnt seems like a long timespan.
You sound much like software engineers sounded when the first coding agents dropped.
"You can't trust the output, it doesn't understand bigger systems"
"Its an art, you need to learn the ropes of it to truly write good code"
Its a very dangerous line of thinking. Software engineering will never be the same, as writing code has basically vanished from the daily workflow. Not for every specialized usecase, but for many.
While I very much see the problem space you're trying to solve, I also like to flag an inherent problem here:
I don't think process and tooling is really the problem. The disfunctional nature of many companies is mainly caused by misaligned incentives and poor leadership, not due to too many tools and disjointed project management.
Unfortunately, while I think the tool looks great, it won't fix the problem for organisations that have it.
It sometimes feels same as with the models, especially in corporate:
- Lots of Haiku around, many mistakes unless process is very clear
- Some Sonnets, still do mistakes but can adapt
- Some Opus, able to improvise and think outside the box.
But even the Human Opus/Mythos are hilariously wrong sometimes.
This looks really cool. Did you think about including automatic worktree creation + sandboxing?
I've built sth similar (more focus on the project setup and being able to work on multiple things at once with a single agent), that uses git worktrees to create a separate space (symlinks .env files) and bubblewrap to isolate the worktree for the agent.