Agentic SDLC Harness & Self-Hosted Infrastructure

Most AI engineering demonstrations stop at "I prompted a model." This project is about the layer underneath: the harness that turns a spec file into a reviewed, committed, production-merged feature — reliably, repeatably, and with a human in the loop at each gate. The SDLC pipeline is a set of structured Claude Code slash commands that drive spec work through a defined sequence: plan → breakdown → implement → test → review → document → wrap-up. Each phase runs in an isolated git worktree, writes a predictably-named report, and gates the next phase on passing validation. Parallel tasks run in separate worktrees and merge in dependency order. The pipeline has been running against real production specs — it is not a demo. The self-hosting layer pairs with the pipeline: a Mac Mini acts as a two-face server. Cloudflare Tunnel exposes the Next.js portfolio publicly at learn-agentic-ai.com without opening a port on the home network. Tailscale provides a private overlay network for local development, internal tooling, and admin access. The site runs on Node.js 22 behind the tunnel; builds are triggered manually and the process restarted in-place — no Kubernetes, no cloud spend, no operational complexity beyond what is needed. The design intent is simplicity at each layer: 18 composable commands rather than a monolithic orchestrator, git worktrees rather than a custom job queue, Cloudflare Tunnel rather than a load balancer. Harness engineering is about making the routine reliable so that the interesting work — the actual AI engineering — can move fast.