One dev + Claude Code agent fleet, running a SaaS. Interactive map of the full agentic SDLC - roast me

I have been building and operating a small production SaaS as a solo engineer for awhile. Over the last months the delivery process moved almost entirely to a fleet of Claude Code agents. I mapped the whole thing (classic SDLC + some dev setup stuff) into one interactive artifact page as a top-level overview for my agentic workflow, see the link in the comment What you are looking at: an agentic SDLC loop (backlog, refinement, design, implementation, verification, human merge gate, deployment, monitoring) plus a "workshop" in the center, which is the stuff the fleet itself runs on (persistent memory with an invalidation doctrine, a CLI token filter, durable tmux sessions, session handover, layered CLAUDE.md context). Every station opens a table of contents. Drilling in shows the skill, hook or systemd unit that actually runs there, quoted from the live setup, plus one real episode from its history. Happy to answer questions about any station. The setup is opinionated in places (hard role boundaries for subagents, deterministic guardrails over prompt-level politeness), so tell me where you disagree.