Set up a repository

Set up a repository

After the machine install, each repository you want PA3 in is set up once. The setup copies PA3's project files into the repository and records them in one commit; it never pushes. There are three ways to run it, and all three do the same install.

Setting up a repository from inside Claude Code: the line PA3 shows at startup, the install and its commit, and the reminder to restart.

Inside Claude Code (the easiest)

1. Open Claude Code in the repository (claude in its folder). Until the repository is set up, PA3 shows this at every start, with your paths filled in:

   PA3: this repository is not set up for Project Architect yet. To set it up, type:
     ! <python> "<pa3-src>/project-architect-3.0/pa_install.py" --project "<your repository>" --yes
   then restart Claude Code (/exit, then claude). Not for this repository? Type:
     ! <python> "<pa3-src>/project-architect-3.0/pa_install.py" --ignore "<your repository>"
   

2. Type the first ! line, or copy it from the message. A line that starts with ! runs straight from your prompt, without a permission prompt, and its output lands in the conversation.

3. When it has finished, restart Claude Code: /exit, then claude. The router only takes over a session when the session starts.

This way asks nothing: it takes every default. The second line tells PA3 to leave the repository alone, and it never offers again there. To turn the offer off on this machine altogether, set "setup_offer": "off" in ~/.claude/usage-ledger/config.json.

By double-click (Windows)

Double-click setup-project.cmd, beside install.cmd in the project-architect-3.0 folder, and pick the repository's folder, or drop the folder onto it. The window shows the install and stays open at the end so you can read it.

From a terminal

Run python project-architect-3.0/pa_install.py --project path/to/your/repo from the folder you downloaded. Without a folder it asks which one, the current folder by default.

These two ways ask a few questions first: one line describing the project, a short interview for the project's card (your name, experience, domain, how much autonomy you want, how PA3 should notify you, your working rhythm and your plan), and a confirmation. From the ! line the card takes the defaults instead: your git user.name, the folder name as the domain, advanced, full autonomy inside an approved plan, and toasts. Edit the ## Developer section of HOW_WE_WORK.md to change them.

What the setup writes

It prints an inventory of the repository first, then:

• PA3's agents, the project-architect skill and the slash commands into .claude/, its tools into tools/, and the templates and 32 seed rules;
• a short CLAUDE.md (an existing one moves to docs/retired/), HOW_WE_WORK.md (the project's card of standing facts), the phase-ends/ folders for plans, task summaries and logs, .claude/pa.json (PA3's switches for the repository, such as the toast level) and a .gitignore block for its scratch folder, .run/;
• .claude/settings.json: the router as the session's agent, permissions for PA3's tools, and auto-compact, recaps and the Task tools turned off;
• a link from Claude Code's memory folder for the project into the repository, so the project's memory lives with its code.

All of it goes into one commit, chore: install Project Architect 3.0. Running the setup again later is harmless: it updates PA3's own files and keeps your edits (Updating).

A 2.0 project

The same command migrates a Project Architect 2.0 repository. Run it with --dry-run first to see every move; INSTALL.md has the details.

Next

Open claude in the repository. The router works out where the project stands and takes it from there; for a new project, that starts with the generation plan (A day in the life).