pm-operating-system

An AI-native product-management OS for Claude Code: a set of PM skills (PRDs, opportunity scoring, market scans, decision logging, weekly reviews) and chat-persona agents (startup-advisor, product-sculptor, growth-engineer, systems-architect) that read and write structured markdown in your product repo. One product per repo — the repo IS the product.

This repo is both the canonical source of pm-os and a working example consumer repo, so you can try every skill end-to-end without leaving it.

Architecture

The repo carries two systems in one tree, separated by audience, and a routing layer on top.

Two modes of work. PM workflows invoke a skill (strategy, discovery, GTM, cadence) that reads and writes structured markdown in context/ + tasks/. Product development follows docs/ and is policed by the constraint layer. Agents are chat personas for pushback — they don't orchestrate, hydrate, or write files; skills own all methodology.

**Why .claude/templates/ lives where it does.** pm-os installs into a product repo as .claude/ (submodule or copied directory). Anything a downstream repo needs at scaffold time must therefore sit inside .claude/ — so the constraint-layer seed files live at .claude/templates/constraint-layer/, not at the repo root.