WrongStack

The deciding trait is that nothing here is borrowed. The kernel, the tools, the permission policy and the memory are all this project's own, which means the pieces fit together properly instead of being three tools taped at the edges. When it works, that coherence is something you can feel in a long session.

It is also an enormous amount of software written by a small project, and you will find rough edges. Pick it if you want one thing that does everything. Pick something small if you would rather it did less and did it reliably.

Writing everything yourself means inheriting every bug the category already found. Provider transports, streaming recovery, partial tool output and interrupted edits are each a long tail of failures the established tools discovered slowly and in public, and none of that learning transfers to a fresh implementation. Seventy-seven declared roles is breadth published as depth; no small project has exercised that many paths.

What it does right is escalate rather than assume. A policy layer that can refuse or ask is the correct default in front of this much reach.

1. Requiring a verification step before a task may advance places the check in control flow rather than in a prompt, which is the difference between a rule and a suggestion. 2. Classifying decisions into three outcomes rather than two is the useful refinement, because escalation preserves the case a binary gate has to guess at.

2. Neither mechanism has a published false-positive or false-negative rate, and for a policy layer those numbers are the entire question. The architecture is stated carefully and its behaviour is uncharacterised.

The effort here is far larger than the attention it has attracted, which is the shape of a project driven by conviction rather than by demand. That produces remarkable software and a difficult sustainability question, because the maintenance load of owning every layer scales with the ambition while the contributor pool stays at whoever showed up.

Moat: none commercial; the asset is one person's willingness to keep going. Likely path: it remains a magnificent personal system, or the pace becomes unsustainable. Position: enjoy it, pin the version, and do not make it the only way your team works.

A view across machines is the feature I have been asking every vendor for. Seeing which agents are running, on whose hardware, against which repositories and at what cost turns a collection of individual habits into something I can actually manage, and per-agent spend limits mean the ceiling is set before the invoice rather than discovered on it.

Identity is still absent. No single sign-on, no provisioning and no audit trail, so the console shows what is happening without proving who authorised it. Approved with conditions: one team, budgets set centrally, and a review before it spreads.

An offline boot flag is a promise nobody else on this board makes in writing. It means the thing genuinely runs with no network, rather than running until some optional feature quietly reaches out, and the fact that the protocol client is one of the features it disables tells me somebody thought about what offline actually requires.

Three local serving stacks have one-command presets and any base URL works besides, so the weights stay on my hardware. Permissive licence on a codebase that owes nothing to anyone else, which is the cleanest fork position here.