TruPath Labs is a publication for operators running real businesses on Claude Code who want to know how someone else does it. The artifacts here are extracted from live work, not theorized.
Three ventures, one stack, biweekly Field Notes from the build. Free MIT-licensed bundles you can drop in your repo. A 26-lesson curriculum that teaches the disciplines, the diagnostics, and the prompt patterns that hold under load.
Most “Claude Code” content online is demo-shaped — toy projects, hello-world skills, outdated by the time it’s indexed. The Labs material is the opposite: tools and patterns running in production today, with the postmortems behind them, the diagnostic prompts that earned their place, and the structural fixes that ship vs ones that don’t.
Three production-tested artifact sets currently public on GitHub:
More bundles ship as the lessons that pair with them go live.
Tuesday morning. Five archetypes rotate based on what actually happened the prior two weeks:
Cross-posted to r/RealityEngineered.
26 lessons across three tiers — Foundations (entry-level), Operating (the core disciplines), Expert (retrofit + scale). Each lesson teaches:
Everything Labs publishes is extracted from a real operating stack across three ventures:
| Editor | Obsidian (vault as the project memory layer) |
| Agent harness | Claude Code (CLI + MCP servers) |
| Agents | Six specialists across three ventures (the routing model in Operating L10) |
| Hosting | Netlify (auto-deploy from main on this site) |
| Site | Next.js 16 + Tailwind v4 + WebGL shader |
| Kanban + tasks | Supabase-backed (cross-venture board read/written by agents) |
| Newsletter | Substack (qcaddy.substack.com) |
| Operating bot | Custom Telegram bot for status, lessons, and conversational continuation |
Pre-launch. SBA underwriting in progress. Ops, vendors, leases, grants. Operating discipline is what makes pre-launch tractable for a solo CEO.
Phase 0 MVP. CV pipeline + ESP32 firmware + smart-board hardware + Mission Control web app. The engineering testbed for everything we write about.
Phase 0 research arm. Sign-language recognition. Public Kaggle notebooks, hardest reasoning problems.
Michael Brenneman. CEO of TruPath Ventures. Stanley, North Carolina. Mostly building, occasionally writing, always behind on email.
My co-founder on Mile High Golf runs his own AI workflow on Codex and ChatGPT. We compare notes constantly. Different agent harness, similar discipline.
Pick the most relevant from the Playbook. Drop it in your repo. ~30 minutes to first value.
Start with Foundations 01 if you’re new to Claude Code, Operating 01 if you’re already running things.
Field Notes deliver biweekly. Free. Unsubscribe with one click. No drip sequence.
If you adopt a bundle and have feedback — what worked, what didn’t, what your domain forced you to change — file it on the GitHub repo. The READMEs evolve with the people running them.
The Labs material is meant to be useful, not authoritative. If something here is wrong for your domain, it’s wrong for your domain — adapt or skip it. The discipline is the point; the specific implementation is up for grabs.