Agentic Engineering — Index

A structured reference to every methodology file in this repo. Start wherever makes sense for you, or follow the suggested order below.

Suggested Reading Order

1. Agent Instructions Starter

Start here. A template that gives your AI project context, communication style, boundaries, and workflow rules from the first message. Copy it, fill in your stack details, and point your AI at it.

2. Escalation Levels

Define when your AI should proceed autonomously (Level 0), inform you after acting (Level 1), propose and wait for approval (Level 2), or stop completely (Level 3). This prevents your AI from making high-risk decisions without oversight.

3. Custom Commands & Memory

End the cold start problem. Set up start session / end session commands and a memory directory so your AI learns from previous sessions instead of starting from zero every time.

4. Shared Context

Commit your AI’s memory to the repo. Every bug someone catches makes everyone’s AI smarter. Every architectural decision is context, not tribal knowledge. The repo becomes the handoff.

All Files

File Purpose Complexity
agent-instructions-starter.md Context and boundaries for your AI Low — copy and customize
escalation-levels.md Risk-based autonomy levels Low — customize examples to your domain
custom-commands-memory.md Persistent memory across sessions Medium — requires directory setup
shared-context.md Team-wide AI memory via version control Medium — builds on custom commands
CLAUDE-template.md Entry point for Claude Code Low — copy to CLAUDE.md

How to Use These Together

Set up a CLAUDE.md and .agent/ directory in your project root:

project-root/
├── CLAUDE.md                ← entry point, auto-loaded by Claude Code
└── .agent/
    ├── instructions.md      ← from agent-instructions-starter.md
    ├── escalation-levels.md ← copied directly
    ├── commands.md          ← from custom-commands-memory.md
    ├── context/             ← from shared-context.md (committed, shared)
    └── memory/
        ├── anti-patterns.md
        ├── context.md
        └── journal.md

Start with CLAUDE-template.md for your CLAUDE.md — it’s auto-loaded by Claude Code and points to .agent/. Other tools (Cursor, Copilot) have their own entry points — see agent-instructions-starter.md for tool-specific setup.

The files reference each other but work independently — use one or use all.

More files are being added regularly. Follow Mats Ljunggren on LinkedIn for context and war stories behind each one.