Set up P-Stack

Configure one repository for the user's selected agent harnesses. The generated model registry is shared. Harness-specific files are adapters derived from it.

Read the harness reference before asking questions or writing files. Run scripts/configure.py --help before using the renderer.

1. Choose harnesses

Ask which harnesses the user wants to configure. The supported values are pi, codex, cursor, opencode, and claude.

Do not create .pi/APPEND_SYSTEM.md, a Codex instruction import, Cursor rules, or OpenCode instruction entries. Pi, Codex, Cursor, and OpenCode read root AGENTS.md directly. Claude Code is the only listed harness that needs a root instruction adapter.

2. Resolve model tiers

For each selected harness, ask for three model choices:

• small for fast, bounded work.
• medium for ordinary implementation and investigation.
• large for design, judgment, and difficult work.