Getting Started

Introduction

PP Core (Pareto Platform Core) is a spec-driven workflow engine that keeps AI coding agents productive across long-running work. It layers structure, memory, and verification on top of the AI runtime you already use, so building software stops being a single long chat and becomes a series of well-defined, repeatable phases.

What PP Core Is

PP Core is a lightweight framework for context engineering and spec-driven development. It does not replace your AI coding tool. Instead, it sits on top of it, supplying the prompts, conventions, and persistent artifacts that turn an open-ended assistant into a disciplined collaborator. You drive the whole flow through slash commands such as /pp-new-project, /pp-plan-phase, and /pp-execute-phase, and PP Core handles the bookkeeping that normally falls through the cracks.

The core idea is simple: write down what you intend to build before you build it, record the decisions you make along the way, and check the result against those decisions when you are done. Everything PP Core does serves that loop.

The Problem It Solves

Most AI coding sessions run into the same three failures, and PP Core is designed around fixing each one.

• Context rot. A conversation accumulates every message, every detour, and every dead end. As the window fills, the model's attention degrades and its answers drift. PP Core counters this by running fresh-context subagents and execution waves, each starting with a clean window of up to 200K tokens, so work is done against a focused brief rather than a cluttered transcript.
• No memory between sessions. Close the chat and the next one starts from zero. Yesterday's architectural choices, naming conventions, and trade-offs are gone. PP Core writes durable artifacts to a .planning/ directory that survive session boundaries and context resets, so the next session can pick up exactly where the last one left off.
• No verification. Many AI workflows declare victory at "execution complete" without ever confirming the output matches the requirement. PP Core makes verification a first-class step, comparing what was built against the spec that defined it.

This is the difference between spec-driven development and "vibe-coding." Vibe-coding asks the model to improvise toward a vague goal and hopes the result is close enough. Spec-driven development pins down the intent first, then holds the implementation accountable to it.

The Phase-Based Model

PP Core organizes work into phases, and every phase moves through the same five steps:

1. Discuss — gather context and surface decisions through adaptive questioning.
2. Plan — produce a detailed, executable plan for the phase.
3. Execute — implement the plan, parallelized into waves where possible.
4. Verify — confirm the built work satisfies the requirements.
5. Ship — package the result and prepare it for merge.

The loop is repeatable. A project is just a roadmap of phases, each run through discuss → plan → execute → verify → ship. Because the state of every phase is written to disk, you can stop after any step and resume later without losing the thread.

Where It Runs

PP Core adapts to a wide range of AI coding runtimes at install time, with no manual configuration needed. Supported environments include:

Command syntax varies slightly by runtime, but the workflow is identical everywhere.

Who It Is For

PP Core is built for developers and teams who use AI coding agents for real, multi-session projects and want predictable, reviewable output rather than one-off scripts. If you have ever lost track of why an agent made a decision, watched quality decay as a chat grew long, or shipped code that no one checked against the original ask, PP Core is designed for you.