tizerluo

The distinction between a procedural graph and a static workflow seems important: self-editing topology can capture reusable strategy, but it also makes regressions harder to localize. I’d be curious whether the refinement loop treats a successful trajectory as sufficient evidence, or uses counterexamples and held-out tasks to avoid encoding a brittle shortcut. A practical evaluation might report graph churn and rollback frequency alongside task success, since a graph that keeps growing could be trading inference cost and auditability for a small gain. The explicit entity–relation–procedure representation also seems like a promising place to attach permissions or provenance to tool calls.

In fact, I don't understand what is this feature‘s difference between CMUX & Orca?

Two things I would want to know before pointing this at a hot service: (1) the overhead budget — when a probe lands on a hot path, is capture sampled or capped per hit, and what p99 latency delta have you measured under load? (2) failure isolation — if probe evaluation itself throws (weird object shape, getter with side effects, huge captured value to serialize), is it contained so it cannot take down the request it is observing? In-process agents live or die by staying boring under worst-case conditions.