Agent OS builder

Build your Agent OS

Not one clever prompt. A small operating system of named, specialized agents with memory, rituals, and compounding learning. Answer a short interview; leave with a personalized bootstrap prompt for Claude Code.

1
Interview before build.It never assumes your role, stack, or standards. It asks, waits, then acts.
2
RULE 0 — never fabricate.Every fact comes from a tool that ran, a file that was read, or something you said. Built into every agent, un-waivable.
3
Contracts before agents.The standards get written first; agents refuse to work without them.
4
Gates are real stops.It builds incrementally and waits for your approval at every gate.

You & your discipline

The operator. Everything else adapts to this.

This shapes the rules, contracts, and agents.
Company / product, and the surfaces you own.

What you ship

2–3 areas you own end-to-end. These become your first named specialists.

The one horizontal that pressure-tests everything. It auto-pairs on every piece of work.

House rules

The non-negotiables your agents never break. Sensible defaults for your discipline — adjust freely.

RULE 0 — never fabricate. Pinned to the top of your CLAUDE.md and inherited by every agent. It can't be unchecked, ever.

Tools & rituals

Where work comes from, where truth lives, and how you give feedback. The OS wires these via MCP later, one by one, verified.

Where the "real" artifact lives.
This shapes the rituals: standups, 1:1s, debriefs.

Here's the OS we'll scaffold

Your answers, reflected back. The builder stops at every gate — this is the first one.

    
        

    Your Agent OS is ready

    One personalized bootstrap prompt. Paste it into Claude Code as your first message — it interviews from your answers, scaffolds the OS, and stops at every gate.

    
          
    How to use it (5 minutes)
    1. Make an empty folder, e.g. ~/agent-os/, and open it in Claude Code (VS Code extension, the app, or cd into it and run claude).
    2. Paste this prompt as your first message.
    3. It builds incrementally and stops for your approval at each gate — don't let it run ahead.
    4. First useful output lands in your first session. It gets sharper every week after.
    Copied to clipboard