Learn
This is the curriculum. Read it in order. Each page says when I last checked the tool claims, and I drop anything I cannot open a live source for.
Start here
Core
Your AGENTS.md is too long
How to write a lean AGENTS.md that agents follow, with a copy-paste template and the docs/ structure that makes it work.
Verified Sep 7, 2026
One AGENTS.md, every tool
You don't need separate config files for Claude Code, Cursor, and Codex. One AGENTS.md works across all three, with a single one-line bridge for Claude Code.
Verified Sep 7, 2026
The 2-4-2 Pattern for Cursor Rules
Stop writing 500-line rule files. Structure your .cursor/rules/ directory with 2 always-on files, 4 auto-attached files, and 2 agent-requested files.
Verified Sep 7, 2026
The Agent Skills standard: one config format for every editor
SKILL.md files work across Claude Code, Cursor, Codex, VS Code, Gemini CLI, and 70+ more tools. Here's how the format works and how to write one.
Verified Sep 7, 2026
Verification loops
Make the agent prove the work with tests, a build gate, and a hook that sends it back when the gate fails.
Verified Sep 7, 2026
Advanced
Multi-model routing: smart agents, cheap execution
Stop paying Opus prices for simple refactors. Route tasks to the right model based on complexity, and cut your AI costs without losing quality where it matters.
Verified Sep 7, 2026
Permissions and unattended runs
The permission modes in Claude Code, Codex, and Cursor, and the smallest safe config for a run with no human at the keyboard.
Verified Sep 7, 2026