Agentic Shell - cli agent adaption layer

Hey folks, TLDR: Spent the today writing an adaptation of cli-agent shell requests, having coded the same across multiple agents on several other projects and open sourced it . So since the advent of autoresearch and, to be frank, wayyy before then, when Tyson Fury taught many of us how to use coding agents , many of us have been experimenting with ways of running coding agent harnesses in deterministic frameworks, moving beyond that of Wiggium Loop itself: A wild wiggium appears while true ; do cat prompt.md | claude -p done As part of my own fork of autoresearch, I put a more deterministic wrapper around it, removing the ability for the agent to decide whether or not to commit, and instead having it based on the output of model training and validation phases. I should note here that Andrej has stated he prefers to not do this with his own autoresearch, I saw a tweet on this and also heard him mention he prefers an interactive approach during his interview with Sarah Guo If we take a