AI Coding Playbook — AI skill for Claude Code

AI Coding Playbook


[](https://github.com/marloncodess/ai-coding-playbook/actions/workflows/selftest.yml) [](LICENSE) 

**Build with AI at full speed – without your agents breaking production.**

This playbook comes out of **3,500+ hours of building real software with AI coding agents** and **millions of tokens** of prompts, sessions, bugs, outages and fixes. It's not theory. Every rule in here exists because something broke without it.

The short version of everything we learned:

**AI agents write code faster than any human can review it. So stop reviewing harder – make the system enforce correctness.** Deterministic engineering governs probabilistic intelligence: in your product *and* in the agents that build it.

One prompt and your coding agent sets your project up to work this way – rules, workflows, guardrails, quality gates and project memory, adapted to your stack.

---

Set up in 60 seconds

Open your project in your coding agent and paste this:

Set up this project with the AI Coding Playbook.
Run: git clone --depth 1 https://github.com/marloncodess/ai-coding-playbook.git .playbook-tmp
Then read .playbook-tmp/SETUP.md and follow every step exactly.

Works the same in **Claude Code**, **Codex**, **Cursor** and any agent that can run shell commands.

Your agent will:

1. Analyze your project – stack, tests, CI, data, side effects, existing agent rules.
2. Install the playbook files without overwriting anything – existing files are merged, not replaced.
3. Adapt the CI quality gates and guardrails to your stack.
4. Draft your SPEC.md, `RUNBOOK.