Who it suits is narrow, and I would rather say so than oversell the audience. It fits someone who already plans in the Claude.ai web chat and wants the execution to happen from there without leaving the conversation. Someone comfortable running their own infrastructure. Someone driving several repos across a host and a VM. It is not for anyone who wants a turnkey product, or pure cloud with no setup.
Why it is best for me specifically comes down to a few honest things. I plan and think in the chat thread, so the copy-paste out to a terminal was the actual friction, and this removes exactly that and nothing else. I already run a VM in a mode with no approval prompts, so headless Claude Code with no taps to confirm fits my setup rather than fighting it. I already keep durable memory in my own notes app, so the context layer the whole thing leans on was sitting there ready. And I like being able to hand a project off between threads and have the new one pick the system straight back up from memory.
The counterweight, which I will not skip, because it is what makes any of this credible: it is fragile. One hung Claude Code run and the whole thing can become a mess. It is a collection of parts held together with polling, and at its worst it has genuinely felt worse than just copying and pasting between the chat and the terminal myself. There were nights of stalled jobs and wrong statuses where that was plainly true. The self-healing work, the silence timeout and the heartbeat reaper, pulled it back from that, and it is now more useful to me than the manual way was. But it earns the word "mostly", not "always", and I am not going to dress that up.
So that is what it is. A personal tool, documented and shared in case the shape is useful to someone, not a product. If one idea from it travels, let it be the context-layer one: teach Claude.ai to operate your tools through durable memory it reads, and a turn-based chat can drive real work on its own.