Actionable methodology files for human-AI collaboration. Feed these to your AI and see what happens.
Most teams using AI have no system — just ad-hoc prompting and hope. These files give your AI the operational infrastructure it's missing: escalation rules, memory, decision boundaries, and guardrails.
Each file is self-contained. Copy it into your project, point your AI at it, and start working.
- Browse the Index to see all available files
- Pick the ones relevant to your problem
- Copy them into your project (e.g., in a
.agent/directory) - Tell your AI: "Follow the instructions in .agent/[filename]"
- Customize the examples to match your project
- When something goes wrong, update the file — that's the methodology
| File | What It Does |
|---|---|
| Agent Instructions Starter | Give your AI context and boundaries from the first message |
| Custom Commands & Memory | Session commands, memory files — end the cold start |
| Escalation Levels | Defines when AI should proceed, inform, propose, or stop |
| Shared Context | Commit AI memory to the repo — team-wide compounding |
| CLAUDE-template.md | Entry point for Claude Code — copy to CLAUDE.md |
See index.md for a structured overview with descriptions and suggested reading order.
More files shipping weekly. Follow Mats Ljunggren on LinkedIn for context and war stories behind each one.
Visit the site: ljunggren.github.io/agentic-engineering
Built by Mats Ljunggren from daily production use across 17 repos. 22 years of enterprise IT, now focused on making human-AI collaboration reliable, repeatable, and safe.
This isn't theory. Every file here comes from real work — bugs caught, incidents prevented, methodology refined through actual shipping.