playnfrontierAI

playnfrontierAI

A shippable skeleton of an agent memory + operating system for OMP — built on one idea: facts are the truth, the card is the window, close-out is one command. Watch it being made: the first boot of this folder — a brand-new agent waking up blind, verifying its machine, fixing a bug in its own gates, all green in ONE SHOT — was recorded live: playnfrontierAI — first boot

This folder is not an app. It is a structure your OMP agent wakes up inside, reads, and builds itself around. Everything here is protocol and structure — you bring the hardware, the model, and the mission.

v3 (2026-09-20): hardened gates (a prose-counts-vs-disk tripwire in doctor.sh, placeholder-path filtering, wrapped-line-safe inbox drain) and the memory sky — tools/sky.sh renders the canon link-graph as a night sky you can open in a browser. The first-boot video below is from v1; the ritual it shows is unchanged.

Boot (the agent's first move)

Open OMP in this folder. The agent's ritual is printed in every load-bearing file:

1. read core/card.md — one screen: live facts, protocols, decisions, watch items.
2. The card shows candidate facts (hardware, engines) — the agent verifies its own machine and promotes them to live. That is the first act of ownership.
3. Work. Capture mid-session into the owning file. Finish with core/tools/close.sh — health → gates → card rebuild → briefing → git snapshot. Zero drift possible.

Quick start (the human)

cd playnfrontierAI
omp                 # continue an existing agent identity
omp --profile me    # isolated profile: blank memory, own sessions/caches (fresh start)

No engine is required to start. If you run a local engine later, point the close-out health gate at it: export OMN_ENGINE_URL=http://127.0.0.1:PORT/health.

What is inside

playnfrontierAI/
├── README.md          this file — the entry point
├── core/              the OMN memory system — canon
│   ├── card.md        generated window — never hand-edit
│   ├── facts/         verified truth (status: live) + candidates to verify
│   ├── decisions/     ADRs — proposed → accepted → superseded
│   ├── projects/      active only; done → archive/
│   ├── sessions/      one daily briefing per day
│   ├── watch.md       candidates & gates, one table
│   ├── learnings.md   append-only SUCCESS log, newest first
│   ├── backlog.md     open work
│   ├── inbox.md       drained by close-out
│   ├── archive/       tombstones — move, never delete
│   └── tools/         build · check · doctor · health · close · recall · graph · sky (the gates)
└── .omp/skills/       the operating skills (see below)

The skills

9 project skills ship in .omp/skills/ — two from this project, seven from obra/superpowers (MIT), the working discipline:

Principles baked into the gates

• Facts are the only prose. Edit fact/decision/project files; README and card are pointers.
• The card is the window. Regenerated by tools/build.sh; staleness shows as ⚠STALE.
• Two freshness gates. check.sh (schema, staleness, dead links) + doctor.sh (reality drift: referenced paths, providers, role pins must exist).
• Trash-principle. Never delete irreversibly — move to archive/.
• Current system is top priority. New artifacts are candidates until 100% verified.

What this is not

No model weights, no API keys, no one's private memory or business data. The gates (check.sh, doctor.sh) will tell you if the structure drifts — trust them before you trust the prose.