Request: review of RTK setup for a public benchmark

[kumarganduri]

Hi! I maintain aimpg (github.com/kumarganduri/aimpg), an open-source tool that reruns real past commits in a sandbox to compare AI coding setups; the tests in each repo decide what counts as solved.

I’m preparing a pre-registered study across 8 public open-source repos, and RTK is one of the setups. It measures tokens per task, cost per solved task and solve rate. Before running anything, I’d like you to confirm or correct how we install RTK. Each run gets a fresh Claude Code config folder, and we run:

rtk init --global --auto-patch --no-trust-filters

with RTK 0.51.0. Is that how you’d want RTK set up for a fair test? Anything we should change?

All raw results will be published so anyone can rerun them, and we’d gladly include your response alongside them. Thanks!