APEX

Agentic Production Engineering eXecution

A cross-platform skill framework for AI-driven replatforming, migration, and high-velocity engineering workflows. Built on the Agent Skills open standard, compatible with Claude Code, OpenAI Codex, Cursor, and OpenClaw.

What Is This?

APEX turns a single engineer into a team lead managing AI agents. Instead of writing code yourself, you decompose features into specs, delegate implementation to autonomous agents at the right trust level, and let self-correcting feedback loops handle the rest.

Inspired by the methodology described in Inside Kilo Speed and the self-improving agent patterns from Addy Osmani.

Skills

| Skill | Purpose | When to Use | |-------|---------|-------------| | apex-decompose | Extract feature specs from existing code | Before any migration or replatforming | | apex-replatform | Implement specs in a new architecture | After decomposition, when rebuilding | | apex-tier1 | Fire-and-forget autonomous execution | Clear tasks with objective pass/fail | | apex-tier2 | Guided execution with checkpoints | Multi-file changes needing steering | | apex-tier3 | Real-time pair programming | Architecture design, critical debugging | | apex-learn | Self-improvement and memory updates | After any task with friction or corrections |

Installation

Claude Code / Cursor / Windsurf

Copy the .agents/skills/ directory into your project root.

OpenAI Codex

Copy the .agents/skills/ directory into your project root. Codex auto-detects skills.

User-Level (All Projects)

Copy to ~/.agents/skills/ for skills available across all your repositories.

Cross-Platform Support

Every skill includes dual scripts:

• scripts/*.sh for Linux and macOS (Bash)
• scripts/*.ps1 for Windows (PowerShell)

The agent auto-detects the host OS and runs the appropriate script.

Quick Start

1. Copy .agents/skills/ into your project
2. Copy AGENTS.md to your project root (or let apex-learn create one)
3. Start with: "Decompose the authentication module from src/auth/"
4. Then: "Replatform this spec into the new Next.js architecture"
5. After any friction: "What did we learn? Update the rules."

The Three Tiers

Choose your level of involvement based on task complexity:

Tier 1 — Fire and Forget: "Add unit tests for UserService." The agent works autonomously and opens a draft PR.

Tier 2 — Guided: "Refactor the payment module to use Stripe v3." The agent works in phases, pausing at checkpoints for your review.

Tier 3 — Pair: "Let's design the new event system architecture together." Real-time collaboration with every decision discussed.

The Self-Improvement Loop

APEX gets smarter over time. The apex-learn skill implements the two-strike rule: if you correct the agent on the same issue twice, it automatically adds a permanent rule to AGENTS.md. This means:

• Session 1: "Use pnpm, not npm." Agent fixes it.
• Session 2: "I said pnpm, not npm." Agent fixes it AND adds - Always use pnpm instead of npm to AGENTS.md.
• Session 3+: The agent never makes this mistake again.

License

MIT