Pi — Minimal Terminal Coding Harness You Own

Practical Notes

• Use AGENTS.md per repo to encode rules (tests, migrations, output format).
• Start with one workflow (lint/test/fix) and add extensions only when needed.

Why a “harness” matters

Most coding agents ship as a fixed product. Pi is intentionally small so you can:

• encode team rules in context files,
• add skills/prompts as reusable building blocks,
• keep the workflow consistent across repos.

Recommended setup for teams

1) Add project instructions

Create AGENTS.md in the repo root:

- Run `npm test` after code changes.
- Do not modify production migrations.
- Keep PR summaries under 10 lines.

2) Teach Pi your repo’s “definition of done”

Ask Pi to:

1. find the relevant scripts (package.json, Makefile, etc.),
2. run checks,
3. propose the smallest diff,
4. re-run checks.

One-shot mode for automation

Pi supports non-interactive prompts:

pi -p "Summarize this repo and tell me how to run checks."

This is useful for CI assistants or scheduled hygiene tasks.

FAQ

Q: Is Pi a model? A: No. It’s a harness that connects to providers and exposes tools, skills, and customization.

Q: How do I keep behavior consistent across repos? A: Use shared skills/prompts and repo-local context files (AGENTS.md).

Q: Can I use it non-interactively? A: Yes. Use pi -p "..." for one-shot prompts in scripts or CI.