PAPE Professional Agentic Product Engineering GitHub

Chapter

Start Here

This guide shows how to build the system around a coding agent, so it carries the load instead of you. Agents can now write most of the code, but run one on prompts alone and you become its full-time operator: re-explaining the project, steering every step, reading every line, and still finding drift and bugs later.

What You Gain

  • Time. The agent works through whole tasks without you steering each step, so your hours go to the problem, not to operating the model.
  • Trust. Tests, checks and rules prove the work is done, so you can ship without reading every line.
  • Control. Your standards live in the repo as files, so every run and every teammate follows them, and the setup gets better each time you fix it.

Three Ways to Start

Go as far as you need; each step goes deeper than the one before.

1. Read the guide. Learn the modern practices at your own pace. Start with the TL;DR, find your level in Which Tier Do You Need?, then read that tier. No setup needed.

2. Point your agent at this repo. It interviews you, then teaches what your work needs. Works with Claude Code, Codex or any agent that reads CLAUDE.md or AGENTS.md. Clone the repo and open your agent in the folder:

git clone https://github.com/krivitsky/professional-agentic-product-engineering
cd professional-agentic-product-engineering
claude        # or codex, or your agent of choice

Then ask: "Interview me about my work and my level, then teach me what I need from this guide, one concept at a time." In Claude Code, hi is enough. With your consent it reads your past prompts to tailor the examples, and it checks that each concept stuck before moving on.

3. Install the plugin in the repo you work in. Already shipping a product? Start here: it audits your setup in minutes, then coaches you as you work. In Claude Code, inside your project:

/plugin marketplace add krivitsky/professional-agentic-product-engineering
/plugin install pape@pape
/reload-plugins
/pape:harness-audit

The audit rates your repo on the six tiers and names the one next thing to fix. It works on the repo you already have: there is nothing to migrate, and it only reads, apart from writing its report to harness-audits/. The coach then stays quiet until it spots a learning moment, and drops one nudge linked to the fix.

After the Audit

  1. Ask your agent to fix the top item in the report's backlog.
  2. Run the audit again. Its Trend section shows what moved.
  3. Commit the harness and the harness-audits/ folder, so every teammate and every CI run inherits them.

What You Need

  • Steps 1 and 2: a browser, or any coding agent.
  • Step 3: Claude Code (the plugin uses its format, so Codex users get steps 1 and 2), plus jq on your PATH for the coach.
  • A team: each teammate runs the three install commands once. The harness itself comes with the repo.