How I actually work with coding agents.
I build and ship production software with coding agents (Claude Code, Codex) as my primary interface. This is the curated set of workflows, skills, and the playbook behind that: the operating system I use to get real, reviewed, production-grade code out of agents, not just demos.
Everything here is battle-tested on real products (Delphy, AstraNova, fermartz.com), with honest notes on what worked and what I'd skip. The mistakes are written down on purpose: the "Receipt" and "Failure mode" callouts are the parts you can't get from a tutorial.
| Pillar | What it is | Think of it as |
|---|---|---|
| Playbook | The principles. Why I work the way I do. | the why |
| Workflows | Repeatable, end-to-end recipes. How I run a piece of work. | the how |
| Skills | Loadable SKILL.md tools you can drop into your own agent. |
the what |
flowchart LR
PB["Playbook — the WHY (principles)"]
WF["Workflows — the HOW (recipes)"]
SK["Skills — the WHAT (loadable tools)"]
PB --> WF
WF --> SK
WF -. "applies" .-> PB
SK -. "embodies" .-> PB
The pillars cross-link both directions: every workflow cites the principles it applies and the skills it uses; every skill points back to the principle behind it. Enter from any side.
- You want a tool you can use today -> browse
skills/andworkflows/, copy what's useful. - You want to see how I think (hiring managers, this is you) -> start with the Playbook and the flagship workflow below. The tools are the receipts; the thinking is the point.
The one workflow that captures my whole approach:
Plan → build → review → ship, with a second model as the skeptic.
One model builds, a different model reviews adversarially, and nothing ships until that independent pass approves it. It includes the real loop, a diagram of every file involved, and the lessons that cost me review rounds before I wrote them down.
Each skill is a self-contained SKILL.md. To use one in your own setup, copy its folder into your
project's .claude/skills/ (or ~/.claude/skills/ for all projects), and your agent can invoke it.
Every doc declares how proven it is, so you know what you're getting:
- battle-tested — used repeatedly on shipped products
- working — solid, used at least a few times
- experimental — promising, still shaking out
Growing in public. The Blueprint is live; the rest of the playbook, more workflows, and the skills library are being added piece by piece (same way I build everything).
Built by Fer Martz — engineer building production AI products: agents, the agentic web, and the tooling that drives them. fermartz.com · github.com/fermartz · LinkedIn
MIT. Use anything here. If it saves you a review round, that's the point.