Skip to content

11 FAQ

Jump to bottom
Varietyz edited this page Apr 2, 2026 · 1 revision

❓ FAQ

Do I need Claude Code specifically?

The methodology works best with Claude Code because claude-setup, PAG, and the memory governance system are built for it. It also works with Cursor, Windsurf, and similar CLI-based AI tools with adaptation. It does not work with web-based AI chat (no hooks, no validators, no session governance).

What is PAG?

Pattern Abstract Grammar. A structured instruction language designed from thousands of hours of Claude interactions. It uses uppercase keywords, phased document structure, validation gates, and embedded algorithmic patterns. PAG is how CLAUDE.md policies, agents, workflows, and planning templates are written. See PAG Keywords and PAG Guide.

Do I need to read everything before starting?

No. Start with the methodology README. Follow links as concepts interest you. Copy claude-setup/ into your project when ready. The methodology teaches itself through practice: the AI reads the documents and applies them as you work.

What is the 150-line rule?

Files that stay under 150 lines tend to be easier to review, test, and reason about. When one grows past that, splitting by concern produces modules that are individually comprehensible. It is a constraint that tends to prevent file bloat in AI-generated code.

What is Phase 0?

Infrastructure established before feature development. Validation hooks, CI/CD, configuration management, testing setup. The reasoning: everything built on top depends on these foundations. Setting them up first prevents rework.

What is XYZW?

Four-axis extraction applied to every task:

  • X (components): entities, files, modules referenced
  • Y (relations): dependencies, connections implied
  • Z (evaluation): decision type (architecture, implementation, review, debug, planning)
  • W (propagation): ripple effects, downstream changes

W is what makes it useful. It forces you to consider what changes downstream, not just what you are working on.

What is the MG compiler?

A Node.js tool that validates memory file frontmatter against registered domains, owners, and type profiles. It runs at session start and before commits. If a memory file has an unregistered domain or missing field, the compiler blocks until fixed.

What is "inherited skepticism"?

When an agent creates another agent, the child inherits: a trust anchor (minimal assumptions), verified/refuted claim tracking, safe iteration variables (bounded loops), and tool calibration. The child does not trust its own tools until it has tested them.

Is this a framework I install?

No. It is a set of documents. Nothing to install, no dependencies, no build step. You read it, copy claude-setup/ into your project, and the AI becomes a disciplined collaborator. The methodology governs itself: the same rules it describes are the rules its own documentation follows.

How do I ask a question?

Navigation

🏠 Home

Quick Reference

Getting Started

Links

Clone this wiki locally