Seeded, not blank.

One command. A folder of Markdown files you own.

The fresh install has one PM door, one shared codex, six raw lanes, six compiled type directories, and twelve installed skills. The tree below is a faithful subset of the created content shape.

phronesis-installation/
├── codex/                     shared principles layer, cited from every door
│   ├── INDEX.md
│   ├── seed/                  46 seeded principles; read-only until adopted
│   │   ├── ai-practice/
│   │   ├── judgment/
│   │   ├── leadership/
│   │   └── pm/
│   └── personal/              the principles you promote, once they are yours
├── domains/
│   └── pm/                    the first door of one engine
│       ├── AGENTS.md          how the workspace briefs your AI surface
│       ├── CLAUDE.md          -> AGENTS.md
│       ├── .cursorrules       -> AGENTS.md
│       ├── USER.md
│       ├── MEMORY.md
│       ├── raw/               Layer 1: what you put in
│       │   ├── transcripts/
│       │   ├── threads/
│       │   ├── inbox/
│       │   ├── meetings/
│       │   ├── research/
│       │   └── degraded/
│       ├── compiled/          Layer 2: what the workspace knows now
│       │   ├── people/
│       │   ├── projects/
│       │   ├── decisions/
│       │   ├── commitments/
│       │   ├── insights/
│       │   └── organizations/
│       ├── memory/            working memory, starts empty
│       ├── artifacts/         durable deliverables you export
│       │   └── prd/
│       └── exports/           disposable renders
├── skills/                    the 12 workflows, below
└── .phronesis/                local runtime config and transient state

The tree omits supported-tool setup files and transient runtime subfolders. Those are plumbing, not content. Everything that matters to your work is Markdown.

Twelve readable workflows ship in the door.

Six run in every door; six are the PM configuration. They are Markdown files with frontmatter, written to be read before they are trusted.

prep-extraction

Assembles a cited, read-only pre-session brief so you know what to watch for before the room starts.

extract-gold-dust

Turns a fresh debrief into durable decisions, person context, commitments, and project state.

compile

Turns raw transcripts, notes, and gathered sources into the compiled state the workspace reads first.

recall

Brings relevant compiled state and earned codex principles to the question in front of you.

meeting-prep

Builds a compact meeting brief from what the workspace already knows about the people and projects in it.

prd-draft

Drafts a decision-ready PRD from trusted workspace state and your framing.

The codex starts with 46 principles.

The seed codex spans four real collections: ai-practice, judgment, leadership, and pm. Seeded principles sit quarantined until you adopt one, so seeded taste never masquerades as your own earned judgment. A sample of the titles on the shelf:

Each seed is a short Markdown file: the principle, why it holds, when it applies. You keep the ones that match how you actually work and leave the rest on the shelf.

Everything above is the starting state. You grow it.

Every session adds to raw/, sharpens compiled/, and, when a call is worth keeping, promotes a principle into your own codex. Six weeks in, the workspace knows your work. Six months in, it knows your taste.

And all of it is Markdown on a disk you control. Open the folder in any editor. Point it at a different model tomorrow. Move it to another machine, or another tool, and what you built comes with you. The workspace is the door into the work. It was never the lock on it.