why is it so slow? ways to work with local repo?

[thivux]

hi. thanks for the great work. i saw your demo video on Eddy's app and wanted to try this out.

1. my first impression is that it seems to take the agent much longer to answer a message compared to using the same model, same effort level on claude code. the thinking is also longer, which costs more tokens. is it an expected behavior?
2. as far as i understand, to use this we always have to start a session inside the firstmate/ dir. is there a way for this to work with a local repo that's on a different path than firstmate/?

thanks.

[atqamz]

yeah I experiencing the same, I really like the philosophy of this repo. but the AGENTS.md itself already bloats the session context, and too many clobbering shell script.

I find no-mistake and treehouse really helpful, kudos to Kun for making it.

but this one, I prefer implement this myself with less bloating things like AXIs, just terminal multiplexer (herdr), worktree manager (treehouse), and optional reviewer (no-mistake).

still on development but feel free to check here: https://github.com/atqamz/secondhand