Why this repo matters
No release captured, 7 developer signals, 2 package/install signals
A closed-loop agent architecture where one LLM generates intent, another executes, and a third evaluates reliability — all grounded by real-world feedback. Synthetic user replaces the human prompter; spine controller replaces the human evaluator; external tool execution prevents the loop from collapsing into self-talk. Design-phase. (0 stars, 0 forks, Python, 5 AI signals, 5 developer signals).