Skip to content

getting-started.md

Latest commit

 

History

History
88 lines (62 loc) · 3.26 KB

getting-started.md

File metadata and controls

88 lines (62 loc) · 3.26 KB
Raw

Getting Started

How to adopt the Agent Operating Framework in your own Claude Code projects.

Quick Start (5 minutes)

  1. Copy AGENT_FRAMEWORK.md to your project root as CLAUDE.md
  2. Fill in Section 0 (Project Identity) with your project specifics
  3. Start a Claude Code session — it reads CLAUDE.md automatically

That's it. The framework is now active.

Recommended Adoption Path

Don't adopt everything at once. Start with what hurts most:

If your agent keeps guessing instead of reading:

Copy examples/claude-code-rules/read-before-acting.md to ~/.claude/rules/

If your agent brute-forces failures:

Copy examples/claude-code-rules/read-before-acting.md to ~/.claude/rules/ — the Three-Failure Stop escalation handles this.

If your agent over-engineers or adds unrequested features:

Copy examples/claude-code-rules/scope-discipline.md to ~/.claude/rules/

If sessions start cold with no context:

Copy examples/claude-code-rules/session-lifecycle.md to ~/.claude/rules/

If changes break downstream dependencies:

Copy examples/claude-code-rules/read-before-acting.md to ~/.claude/rules/ — Gate 0 forces reading the resource before touching it; pair with scope-discipline.md Gate 5 (Dormant Code Check) for caller mapping.

Where Rules Live

~/.claude/rules/          ← Global rules (apply to all projects)
.claude/rules/            ← Project rules (apply to this project only)
CLAUDE.md                 ← Project root (loaded every session)

Global rules go in ~/.claude/rules/. Use for universal behavior like "read before acting" that applies everywhere.

Project rules go in .claude/rules/ inside your repo. Use for project-specific constraints like "always use pytest, not unittest."

CLAUDE.md goes in the project root. Use for project identity, output contracts, and high-level guidance.

File Structure

A mature project using this framework:

my-project/
├── CLAUDE.md                    ← Section 0 filled in, points to rules/
├── .claude/
│   └── rules/
│       └── project-specific.md  ← Project-scoped rules
├── src/
└── ...

Your home directory:

~/.claude/
├── rules/
│   ├── read-before-acting.md    ← Universal
│   ├── scope-discipline.md      ← Universal
│   └── session-lifecycle.md     ← Universal
└── projects/
    └── <project>/
        └── memory/
            └── MEMORY.md        ← Persistent memory for this project

Growing the System

  1. Start with CLAUDE.md only. Fill in Section 0. Use the framework as-is.
  2. When a mistake happens, write a memory entry. Soft guidance.
  3. When the same mistake happens again, promote to a rule. Hard constraint.
  4. When a rule gets ignored, promote to a hook. Automated enforcement.

Read guides/enforcement-architecture.md for the full escalation model.

What NOT to Do

  • Don't adopt all 7 sections on day one — start with what hurts
  • Don't write rules for problems you haven't had — rules without incident reports get ignored
  • Don't put everything in CLAUDE.md — it loads every session, keep it lean
  • Don't skip Section 0 — without identity and output contracts, the rest is enforcement without direction